CONTROL MODEL DEVELOPMENT

FOR PACKED BED CHEMICAL REACTORS

Thesis by

Rohi t Khanna

In Partial Jilulfi.Ument of the Requirements

for the Degree of

Doctor of Philosophy

California Institute of Technology

Pasadena. California

1984

(Submitted May 16, 1984)

 0 1984

Rohit Khanna

All Rights Reserved

 iii

To my parents for their support and encouragement

throughout my education and to my wife for her love
and understanding throughout the past six years.
 iv

This thesis would clearly not be possible without the guidance of my advi-

sor, Dr. Seinfeld. 1 am grateful that he was able to allow me to conduct research

in the areas of my interest without significant restrictions. His assistance is

greatly appreciated. I would also like to thank the National Science Foundation

tor their graduate fellowship that provided much of the assistance for this
research.

Much of the real credit for this thesis should go to my family. Without the

visions and foresight of my parents and without their high expectations this

would never have been possible. Most of all I would like to thank Debbie for her

love and support throughout the time at Caltech and especially on all those

nights when she wished that I could be home but understood that I had to work.

Finally, I would like to thank Dave and Barry for helping provide the outside

interests necessary to keep life stimulating and enjoyable and to retain

enthusiasm in the research.

 v

ABSTRACT

Although control algorithms have been conceived for industrial chemical

systems. their acceptance by industry has been slow due to a lack of direct

experimental evidence of their effectiveness and to volumes of conflicting, or at

least incompatible, recommendations on control structure design. This thesis

provides the basis for a concerted theoretical and experimental program in mul-

ti variable process control structure design for packed bed chemical reactors by

presenting an in-depth control analysis of a practical, multivariable, distributed

parameter system-the heat conduction problem defined by the simple diffusion

equation-using both frequency-domain and time-domain analyses and the for-

mulation, numerical solution, and analysis of a detailed model for packed bed

reactors, along with reduction to a low-order state-space representation suitable

for on-line process control.

The study of the heat conduction system allowed for consideration of vari-

ous control design techniques and the relation between measurement structure

and control system design. This study shows that the choice of measurements
and their locations significantly affects the optimal control design and the use-

fulness of the different design techniques and the importance of an accurate

process model and the necessity of model reduction to a low-order state-space

representation for control structure design and implementation.

The second portion of this study provides a detailed mathematical modeling

analysis of packed bed catalytic reactors that significantly extends previous stu-

dies· in the detail of the model and in the consideration of all aspects of the
model development and reduction to a state-space control representation. The
 vi

general view that modeling simplifications are desired since they lead to a

reduction in numerical solution effort is contested, and it is shown that many

simplifications are no longer necessary with today's advanced computational

capabilities. A unified approach to dynamic reactor modeling is developed and

its importance in the accurate description of dynamic and steady state reactor

behavior, in the investigation of reactor start~up or the effects of process distur~

bances, and in lhe development of an accurate reduced stat~space model for

the design of control structures to stabilize the reactor under various distur~

bances or to provide optimal system recovery from input changes is shown.

 vii

TABLE OF CONTENTS

Acknowledgement vi
Abstract v
Notation xi

1. Introduction 1

1. Objectives of the Current Work 2

2. Industrial Control Considerations 5

3. Review of Experimental Control Studies 8

2. llultivariable Control Structure Design for a Heat Conduction System 13

1. Introduction 14

2. Preliminary Analysis 18

1. Model Red.uctiun. 18
2. CantrolZa.lTility 26

3. Root· Locus 27

3. Tim~ Domain Analysis 35

1. qptimal Ccmtrol 36
2. Mod.a.l Cr:mtrol 40

4. Non-Interacting Control 46

1. Prn"Ject .Dynamic Cl>f111>ensa.tion 46

2. Stea.d.y Sta.te .Decrru:pling 49
3. Set .Pm:nt O>mpensa.tion 49
4. Inner-Loop .Decoupli:ng 50
 viii

5. Inverse Nyquist Array 58

6. Characteristic Locus Method 78

1. Ba.sic System Analysis 84

1. O:tse !: Measurements z 1=0.3, z 2=0.7 84

2. Chse II: Measurements z 1=0.4, z 2 =0.6 86
3. Chse III: Measurements 2 1=0.4, z 2 =0.8 87
2. Inner Loop .Analysis 89

1. Qzse I: Measurements Zt=0.3, z 2 =0.7 92

2. Chse II: Measurements z 1 =0.4, z 2 =0.6 94
3. Cb.se III: Measurements~ l;;;:Q.4, ze;;;;0,6 95
3. Modified Inner-Loop Analysis 95

7. Control System Performance 119

1. Case I: Mea.su.rements z 1=0.2, z 2 =0.8 124

2. Ca.se II: Mea.surem.ents z 1 =0.3, z2=0.? 125
3. Case III: Measurements zl=0.4, z2=0.6 126
4. Cb.se IV: Mea.su.rem.ents z 1=0.4, z2=0.6 128

8. Discussion of Control Analysis 161

3. U:athem.aticalllodel of a Packed Bed Catalytic Reactor 166

1. Introduction 167
2. Review or Packed Bed Reactor Modeling 172
3. Model Development for Packed Bed Catalytic Reactors 179
1. Reactor System 179
2. Reacf:ion .Ni:netics 182
3. Formation of OJmplete Mathematical Model 187

4. Mathematical Rela.tianships 192

 ix

5. N'tl.t71.BriDal Solution 199

1. Solution Techniques 200

2. Orthogonal Collocation 204
3. Model Reduction 20f
4. Numerical Sim:ulaticm 216
4. Model Analysis 223

L Modeling Pa.rameters 224

2. Steady-State Beha:uiar 229
3. Dynamic Sim:ula.tian.s 230
4. EJ!ects oJ RtuJ.clrYr Dperati:ng Conditions 232
5. Radial Ccmcentra.ticm Analysis 234
6. Adiabatic Analysis 238
7. ImprYrtance oJ Thermal Well 239
8. Dispersion Effects 241
5. OCFE Model Analysis 271

1. Fcrrmulation of oc;:Ji'B' Technique 272

2. Madel Comparisons 279
6. Control Model Development 293

1. State-Space Representation 293

2. Model Linea:ri.rza.ticm. 296

1. Linearization of the Reaction Rates 297

2. Lin.earizatian of the Al.gellraic Equations 298
3. Linearization of the Differential Equa.titm.s 300
4. Nu:rnerical Sol~ of the Li:n.earisl:ed Model 302
3. State-Space Si.m:ulatians 304
7. Model Dimensionality 317
 X

1. Model Discretization 317

2. Physical Modeli:ng Sim:plifications 320

1. Horrwgeneous Analysis 321

2. Quasi Steady State Approzi:ma.tion 326

3. Negligible EJn.ergy Accumulation in Gas 328
3. Model Reduction 329

1. Davison's Method. 331

2. Marshall 's Method. 334
4. Discussion of Reduced. Model 336

4. Results and Di.scUSBi.on of Future Experimental Studies 351

1. Conclusions of Current Analysis 352

2. Parameter Estimation and Model Verification 361
3. Control Model Configuration 367
4. Control System Design 374

5. Appendices 378

1. Computer Programs for Heat Conduction Problem 379

2. Normalized Packed Bed Reactor Model 385
3. Radial Collocation of Packed Bed Reactor Model 391
4. Computer Model of the Packed Bed Reactor 398
5. Orthogonal Collocation on Finite Elements Model 452
a. Linear Packed Bed Reactor Model 466

6. References 485
 xi

NOTATION
 xii

A state matrix

ft.v.J3v collocation weights for first and second derivative

B control matrix

c concentration, g-moles I cm5

element ij of measurement matrix C

heat capacity, call g °K

c measurement matrix

d disturbance vector

radial collocation weights

D bulk molecular diffusivity, em 2 I sec

activation energy, call g-mole

pressure dependence constants for steam-shift reaction rate

element ij of inner-loop compensator :r

F inner-loop compensator

element of process transfer function Gp

controller transfer function

interaction compensator transfer function

process transfer function

set point compensator transfer function

heat transfer coefficient between phase i and j, cal!sec em aK

heat of reaction, call g-mole

 xiii

molar ftu.x relative to molar average velocity

k thermal conductivity

reaction rate constant

Blake·Kozeny constant

diagonal gain element of K

radial (axial) thermal conductivity, cal/sec em °K

K diagonal proportional gain matrix

methanation reaction rate constants, atm- 1

low and high frequency compensators

equilibrium constants, atm- 2 for methanation

Lagrangian polynomials

L reactor length, ern

controllability matrix

molecular weight of gas, g I g·mole

molar ftu.x with respect to stationary coordinates

p -pressure, atm
total pressure neglecting mole change, atm

closed-loop characteristic polynomial

partial pressure of species i, atm

open-loop characteristic polynomial

Pe dimensionless Peclet number

 xiv

Q open~loop transfer function

r radial coordinate. em

r normalized radial coordinate, r /R 1

radial collocation point

R reaction rate, g~mole/sec cm3

R closed-loop transfer function

radius of thermal well and outer wall, respectively, em

R' normalized reaction rate

Rg universal gas constant, 1.91:i7 cal/g~mole aK

t time. sec

t normalized time for heat conduction problem

T absolute temperature, aK

u control vector

control inputs to heat conduction system

Ug interstitial velocity of gas, em/sec

u internal energy

overall heat transfer coefficient between phase i and j, cal/ sec °K

normalized ftuid velocity, u1/~

v volume, cm3

weighting functions

X state vector
 XV

xi mole fraction of species L g-mole if g-mole total

y normalized temperature distribution for heat conduction problem

y output vector

y0 normalized temperature distribution at t=O

Yd desired normalized temperature distribution

y, normalized mole fraction, Xt.rx&

z axial coordinate, em

z normalized space coordinate for heat conduction problem

z1 • z2 outputs for heat conduction system

 xvi

Greek letters

a dimensionless axial dispersion, oc: Pe; 1

(J dimensionless radial dispersion, oc: Pe;1

6 moles CO reacted in methanation per total inlet moles

void fraction of bed

1'} dimensionless heat capacity

7 dimensionless heat transfer coefficient, oc: St

dimensionless heat generation parameters

"
eigenvalues

Biot numbers

diagonal matrix of eigenvalues

viscosity

constants from radial collocation

normalized radius of therma well, Ro/R 1

dimensionless heats of reaction

p density, g/cm3

a dimensionless reaction coefficients

dimensionless pressure drop, Pz=LIPz=e- 1

normalized time, (t ·il~o)IL

normalized temperature. TIT I.)

normalized axial coordinate, z/L

 xvii

Subscripts and Superscripts

0 value at inlet

b reactor bed

f value at outlet

g gas

h heat

m mass

M methanation reaction

r radial

s solid catalyst

s steam-shift reaction

t thermal well

w cooling wall

z axial

based on inlet conditions

steady state

deviation variable
 Chapter 1

INTRODUCTION
 -2-

One of the major difficulties in the application of advanced process control

theories to industrial chemical processes is the lack of complete understanding

of the chemical process and the inability to predict system behavior in the pres-

ence of unknown and uncharacterized disturbances. This difficulty is com-

pounded by physical and chemical interactions or nonlinear coupling of the pro-

cess variables. Although complex control algorithms have been conceived for

industrial chemical systems, their acceptance by industry has been slow due to

a lack of direct experimental evidence of their effectiveness and to volumes of

conflicting, or at least incompatible, recommendations on control structure

design. Significant research efforts are necessary in duplicating industrial

processes in a research setting and providing a unified approach to the

mathematical modeling and control structure design.

This thesis provides the basis for a concerted theoretical and experimental

program in multivariable process control structure design for packed bed reac-

tor systems. Along with the detailed design and construction of both a kinetics
and pilot-scale control reactor {Strand, 1984), this work presents the necessary

prerequisites to a substantial effort in the study of the applicability of process

control theory to a laboratory- or pilot-scale industrial chemical process. In

view of these objectives, this thesis is divided into two major sections:

- an in-depth analysis of a practical. multivariable, distributed parame-

ter system-the well-defined beat conduction process defined by the

simple diffusion equation-using both frequency-domain and time-

domain analyses and

-the formulation, numerical solution, and analysis of a detailed

dynamic model for packed bed reactors, along with reduction to a low-
 order state-space representation suitable for on-line process control.

The first section of this work, the analysis of the heat conduction problem,

is presented in Chapter 2 and is a self-contained presentation of many of the

control aspects of a multivariable, distributed parameter system (Khanna and

Seinfeld, 1982). Although this heat conduction system is much simpler than the

packed bed reactor, a considerable amount of insight into control structure

design can be obtained from this process since the control aspects are not lost

in the complexity of the mathematical system as they may be in an initial

detailed control study of the experimental packed bed reactor. In particular, an

analysis of a number of multivariable process control strategies, including non-

interacting control, optimal control. inverl:!e Nyquist array, and characlerislic

locus techniques, is carried out theoretically on a one-dimensional. two-input

heat conduction system. The potential improvements in control performance

through the usc of extra measurements and through the appropriate selection

of measurement locations is assessed and a new non-interacting control stra-

tegy, termed inner-loop decoupling, is developed.

The remainder of this thesis centers on the complete dynamic modeling

analysis of a packed bed reactor, along with reduction to an accurate low-order

state-space representation suitable for control studies. According to Jutan et al.

( 1977). this is

"one of the more complicated processes to model in chemi-

cal engineering. Because of this, it is essential when deriving
a process model to keep in mind the purpose for which the
model is to be used. If, for example, the model is to be used
as the basis for on-line regulatory control of the reactor,
large simplifications to most of the models proposed in the
literature must be made. Although most of the models tend
to be somewhat complex and, in general, unsuitable for con-
trol. by examining the formulation of these models some
insight into the important effects occurring within a reactor
may be gained, and ideas for simplifying the models for the
purpose of the control may be found."
 -4-

In view of these comments, this study allows accurate description of dynamic

and steady state reactor behavior for process optimization and design. for the

investigation of reactor start-up or the ef:J'ects of process disturbances, and for

the analysis and design of control structures. Various common assumptions

and model structures are considered, and the appropriate numerical solution

techniques are discussed. This analysis is not intended to be specific to any par-

ticular packed bed reactor system but rather to present a detailed study of

modeling techniques, assumptions, and solutions and to develop a unified

approach to dynamic reactor modeling and control model development. The

work significantly extends previous studies in the detail of the mathematical

model and in the systematic consideration of all aspects of the model develop-

ment and the reduction to a state-space control representation.

 -5-

1.2 INDUSTmAL CONTROL CONSIDERATIONS

Until only recently, the process control industry has been dominated by

applications to mechanical or electrical processes where systems are generally

well-defined and numerically simple due to minimal nonlinearities and relatively

simple physical characteristics. Even so-called 'modern' control theories

developed in the 1960's have only played an important role in fields such as

robotics and the space program. Their applicability to complex industrial chem-
ical processes which are inherently burdened by nonlinearities, large time
delays, and distributed parameter behavior has been extremely limited in spite

of an enormous effort by researchers. Although significant theoretical efforts

have been made by many research groups in the development of control algo-

rithms for ill-defined distributed parameter systems as are common in the

chemical industry, these efforts have not found wide application in actual indus-

trial problems. Thus during the past twenty years, a significant control 'gap' has

developed between process control theory and practice (Foss, 1973; Seborg and

Edgar, 1982).

Due largely to increased costs for energy, increased governmental safety

and environmental restrictions, and increased foreign competition, significant

efforts are now being made by industry to modernize and automate production
facilities with a major emphasis on integrating energy requirements and improv-

ing throughputs and system performance (Ray, 1981). Additionally with the

major improvements and general acceptance of computers suitable for on-line

control, attempts at applying sophisticated control theories are only now begin-

ning. One of the conclusions of the discussion on the problem of the control

'gap' between academic theories and practice is that there exists a ·pressing

need for careful, systematic studies of the design and implementation of control

systems for pilot-scale industrial processes, thus providing the prime motivation
behind the current theoretical and experimental program for which this thesis

provides a basis.

Several steps are necessary prior to developing and testing control stra-

tegies. These include the design and construction of a pilot-scale system pos-

sessing much of the modeling characteristics and control difficulties of the

actual industrial process, the development of an accurate mathematical model

for process design and analysis, and the reduction of the full model to one suit-

able for control structure design and on-line control. The first of these steps

was performed by Strand (1984), and the latter two steps are the concern of this

current work. A packed bed nonadiabatic chemical reactor was selected as the

chemical process to be studied due to its complexity and inherent control

difficulties and its extensive industrial importance for carrying out exothermic,

gas phase reactions.

Upon completion of these preliminary research efforts. the elements of the

control structure:

• the measured variables,

• the manipulated variables,

• the control configuration connecting the measured and manipulated

variables, and

• the control logic governing the behavior of the manipulated variables

can be considered. In particular, it is important to determine control struc-

tures that make optimal use of the availiible measurements in the chemical pro-

cess. Although in general there may be quite a few measured variables available,
they may not be the variables of most concern, and these variables may need to

be reconstructed using the process model. An example of this is the measure-

ment of the outlet gas temperature in the packed bed reactor from which the
 -7-

outlet concentrations may need to be estimated. Further problems with meas-

urements result from various random and systematic errors or noise in the

measurements and long delays due to the complexity of the chemical or physi-

cal analyses. 1 The use of an accurate process model can in many cases minimize
these diffl.culties by allowing prediction of some unavailable measurements, by

improving the knowledge of the system performance through simulations, and

by allowing design improvements to the process.

Thus the central role played by dynamic and steady state models in the
design and optimization of chemical processes and in the development and

application of control strategies justifies considerable effort in their develop-

ment. The philosophy of this modeling work is presented by Foss ( 1973),

'Forms ot the models range from sets of nonlinear

differential equations to empirically or experimentally
derived transfer functions. The forms of the models may
not be selected arbitrarily: they are determined in part by
the control objectives and the type of control analysis to be
pursued. In short, process modeling is a substantial and
crucial task, and by no means routine. .. . The operation of
control systems of modern design also requires estimates of
the process states used for control. This requires a process
modeL perhaps different than that used for design calcula-
tions, and a means of rapid solution of the model equa-
tions."

The work presented in this thesis is intended to minimize the diffl.culties associ-

ated with process modeling by providing an accurate unified approach to the

model development for packed bed chemical reactors.

1. Such as using a chromatograph to measure outlet concentrations :!'rom a reactor.

 -8-

1.5 REVIEW OF EXPERIIIENTAL CONTROL S'I'UDIES

Several experimental laboratory-scale packed bed reactor systems have

been studied in terms of control considerations. Table 1.2-1 outlines some of

the work done by several groups that have made significant contributions in this

area. Their work is by no means the extent of control applications for chemical
processes (there is much published work on the control or separation processes

and other reactor configurations) but is the major extent of published experi-
mental application of control techniques to packed bed reactors.

The Denmark group (Clement and Jorgensen, 1981; Clement et al., 1980;

Hallager and Jorgensen, 1981; Sorensen, 1977; Sorensen et al., 1980) considered

an adiabatic pilot-plant chemical reactor with the reaction between oxygen and

hydrogen over an alumina supported platinum catalyst. Initial dynamic model-

ing and experimental studies were carried out by Hansen and Jorgensen (1974,

1976ab) and by Sorensen (1976). The group considered various reactor and

control models and investigated control strategies based on optimal control.

direct Nyquist arrays, and the self-tuning regulator.

The Berkeley group (Foss et al., 1980; Michelsen et al., 1973; Silva et al.,

1979; Vakil et al., 1973; Wallman et al., 1979) studied the same hydrogen/oxygen

system but used a two-bed reactor structure with an interstage quench stream.

Again the beds were taken as adiabatic, and various control strategies were de-

vised. The first control studies by Silva et al. (1979) and Wallman et al. (1979)

considered the control of the quench fiowrate and temperature using tempera-

ture measurements and a product concentration estimator using the stochastic

linear quadratic regulator and multivariable integral control. The latter work

by Foss et al. (1980) uses the tlowrate and temperature of the quench stream

and the feed temperature to regulate the product concentration and tempera-
 -9-

ture using the characteristic locus method of control system analysis. A major
significance of their work was the consideration of the large number of available

measurements and the appropriate choice of control contlguations.

The McMaster group (Jutan et al., 1977; MacGregor and Wong, 1978; Wright

and Schryer, 1978) considered the hydrogenolysis of butane carried out over a

nickel on silica gel catalyst in a nonadiabatic packed bed reactor. The work by

Jutan et al. ( 1977) provides an excellent foundation for packed bed reactor

modeling and control studies for multiple reaction systems by specifically con-

sidering the state-space model development, the parameter estimation and sto-

chastic disturbance identification, and on-line linear quadratic control. The ear-

lier work by MacGregor and Wong (1978) and Wright and Schryer (1978) deviated

from the mechanistic approach to reactor modeling taken by most studies

where models are developed by careful consideration of the important chemical

and physical phenomena occurring within the process. They considered the use

of statistical methods to identify process transfer functions from empirical

input/ output process data. They then applied a model reference adaptive con-

troller and a state-space stochastic linear quadratic regulator.

Finally. the dynamic behavior of an autothermal reactor with internal

countercurrent heat exchange using the steam-shift reaction was modeled by

Bonvin (1980) and Bonvin et al. (1979, 1980). Modal control using state feed-

back was found appropriate for stabilizing the reactor around an unstable

steady state.
 f)pnMRrk Group

Syat8 leferencu Orleln~l Hodel Cont_':£!._!fodel Control Variables Control Stnteu

2
H + o2 + 2H o Sorensen (1977) -~+--l_U_!t~.o - Cot }:watton Mtaftsured Stochastic linear
2 2 ;l7. PeMl" ;)Z2 C0 (4-A rotnts) - 1l t""'P"rature qu•dratlc
Cleaent. Jorf(l!'nsen,
2 - 1~tne~trized - 6 o concentration reRulator
Suren•n•n (191!0) - ~ +_I_ ~ - ~~ (~-Y ) 2
- Tnt<1l flow-rate
:•~ Perz" az2 r"TR w Hanhmlated
- J-"l'f"tl teaperat1.1f't!
- fl_ fiHR • _;l! - o 2 feed concen-
CT ;Jt tration
0
- Tntnl ftovrate

Cle..nt • JorR.ensen Same 111 s above Snmr AS above s ...._. .as above Direct Nyquist array I
(198!) ~
0

RallaAf!r. Jor~ten­ x(t) • A(q- 1)x(t-1)+8(q-l)u(t-d) - te.1At 8CJUATeR Same AS above Hfl<lll ... u -tuntna
'"" (!'Jill) !dent H lent ton r<Jiulator (LQR)
+e(t)+c - Tf.e series
Model
or

Kl(t) • ·~(t)OI + el(t)

Acllabattc

Table 1.2-1
Summary of Published Research on Packed Bed Reactor Control
 Berke t ey Group

Syate. Reference ortgina I llod•d Control Hodel Control Variables Control Strateu

11 ..0
2 2
+ 2H
2
o Silva, Vallun, rlX RO • c -tlneari7.atlor~ Measurpd Stochastic linear
- ;iZ - co
Fou (1979) -Co II ocRtl"" - 14 te,..,eraturea quadratic
2Nu 2Hu - 14 points ManipulAted re~~tulator
~ • _ _s (ts-Yl + 5
-p- {Yw-Yl -Drdt'r r .. ductlon - {luench flovr•te
- oZ PeTR ~TR - 1 point" - Te•perature at
avs - l!t!thnd of quPnch •lxinA
211u 5 Y) + ...!!._ AIIR

~~
- - (Ys- - -~ D-nvlfaon roint
PeTR CTo
avw
2)'~"w(Y
Pe W
-Y)- yU{YW-YE)
-~

Wall•n, S!lva, Sa.e aa above- Sa• att above Heaaured Stoc:haettc Unear
Fon (1974) - 5 tell!'~ in lied It q~drattc
Quench Man!pu!At..d re11ulator vtth
- Sa.e aa above integral aetlon

....
--- ~ Pose. !d.unda.
Kouvar1taltls
Salle aa above Sn.- as abov~ Me•attred
- 7 te....,raturn
Chauetertatf<: locua
Mthod

- (1980) - F.xl t concentration

Mnntpulltt ..d
- r.,.d te,..,erature
- Quench te....,rature
- Qu•nch flovrate
loth lleda
AUabattc

Table 1.2-1 Continued

 McM~tat~r Group

SysteM Jlef.,r•ncu Original Hodel Control Hodel Control Variableo Control Strategy
I

Jut an • Tre-mb 1-av.

lutan• Hydro11HolyaJa
0 Jndepen4ent H~rr.re~or. Wrt~ht ax 1
t. ~ ( 1 a. J ~~~ 1
~x 1
-Qmtsf !ls .llppro-xi- Hettftur~d Stochaatfc ltnear
( 1977)
- TfZ + PeHRm ;:;R R ;tjt - ~ • ;ll IMtion - 11 adal t•~era- quadrat tc rekulator
Reactions)
-Collocllt ton ttn-e
at k altav} J Ra~id pointe - 1\x!t concentra-
- ~z + Pen• :>R i! :lR 7 axlltt pointe tiona
ay
+-!- E
CT 1•1
""~~ -~
-l.in•Arhed Manipulated
- In I et 11 2 floon:ote
0 - lnl<?t c 11 fl--
6 10
rate

~
MacGregor, Wong wl • ciA + 1 i - 1,2,} -t.tne-ar t'e~re-s­ Sa• ae above State space atochaatic
t " P.t .. t N
(1'178) Rion or cone, II near quadratic
ttt"'''4 ~~~easur~­ r~1uletor
MPnts
-Ttae- •trie-s
analysis (AUMA)
0..

Wright, Schtye1"
(1978)
.. .... ..
X (k+l) • A

y (k) • II x (k)
X (It) + II u (k) -Fo-rw Tl!fe-rence
...,del (2z21
-Tt.e-•erte-tt
SaM ae above Model reference
Adaptive control

ana I yah to eati-

"• • KFB"• (k) + K;pYsp (It) ute- transfer
functt()tt

lfonadta.. t tc

Table L2-1 Continued

 Chapter 2

:MULTIV.ARIABLE CONTROL STRUCTURE DESIGN FOR

A HEAT CONDUCTION SYSTEM

 - 14-

2.1 INTRODUCTION

Classical process control techniques were largely developed on a trial and

error basis, with the theoretical concepts developed later to substantiate the

empirical results. The classical controller was based on a single-input, single-

output (SJSO) system with three types of possible control action-proportionaL

integral, and derivative-based on the feedback error. Although design tech-

niques such as root-loci, Bode diagrams, and Nyquist plots led to empirical rules

for setting the appropriate amounts of control action, these methods were lim-

ited to SISO systems. Since most processes have multiple inputs and outputs,

additional considerations became necessary, due to the failure of single-loop

analysis for interacting loops. Since frequency-domain methods dominated con-

trol system design in the scalar case and led to relatively simple controllers, it is

not surprising that considerable effort went into extending these methods to the

multivariable case. However, direct extension of scalar frequency-domain pro-

cedures was not possible, and major modifications of the existing theories were

necessary to meet the design objectives. Furthermore, design complexities were

often enhanced due to the additional objective of noninteraction. Nevertheless,

several excellent frequency response procedures were developed in the late 60's

and 70's, led by the work of Rosenbrock, MacFarlane, and Kouvaritakis.

At the same time, ''modern" control techniques were developed. These

methods rely on an exact knowledge of the system state, which is reconstructed

from a finite number of measurements using current theories in optimal filter-

ing, smoothing, and estimation. The ideas have been extended to account for

modeling and measurement uncertainties and inaccuracies. Although these

''modern" methods, which rely heavily on variational calculus and dynamic pro-

grarn.ming, generally lead to a more complex control structure, they are less

heuristic than the frequency-domain methods and allow for more precise
 - 15-

control objectives, such as the minimization of the variance in the state vector.

This chapter describes an in-depth analysis of a practical. multivariable,

distributed parameter system using both frequency-domain and time-domain

analysis. As a typical distributed system, a one-dimensional, heat conduction

problem is considered. The process is described by the difiusion equation

tJ2y(z,t) By(z.t)
(2.1-1)
az2 Bt

where the temperature distribution, y(z,t). is dependent on the space coordinate

z. which is normalized (0.0 < z < 1.0) with respect to the thickness of the sys-

tern, and on time, which is normalized so that the coefficient corresponding to

the thermal difiusivity is unity. For simplicity, the controls u 1 (t) and u 2 (t) are

taken to be the heat fiuxes at z ""' 0 and z ""' 1. Thus the initial and boundary

conditions are

y(z,O) =Yo(z) (2. 1-2)

By(z.t)
Clz
j
•=O
= -ut(t) . By(z,t)
az
j =u2(t).
•=1
(2. 1-3)

Although this heat conduction process is a relatively simple control prob-

lem due to the simplicity of the model and ease of obtaining measurements. the

analysis leads to conclusions that can be extended to general, multivariable con-

trol theory. Actually the system is an excellent choice, since Equations (2.1-1) -

(2.1-3) can be solved analytically to give the temperature distribution for any

control action. Thus model reduction and control design techniques can be

applied to the reduced system and compared with the actual process model.

Additionally, the heat conduction system is a highly interacting process with

implicit transportation lags.

Due to the simplicity of the model. both distributed and lumped analyses
 - 16-

can be performed. Since most frequency-response techniques require a low-

order, lumped, state-space representation, model reduction is an important

step in the analysis. Section 2.2 discusses the model lumping and reduction

using exact techniques. Additionally, an analysis of output and system control-

lability, along With a multivariable root-loci analysis, is presented. Much insight

can be obtained from these preltminary considerations.

Section 2.3 discusses the time-domain analysis of both the lumped and dis-

tributed models of the system. Both optimal feedback control and modal tech-

niques are discussed in detail, along With derivations of the control schemes.

Since much work has been published on the application of optimal control

theory to the single-input, one-dimensional heat conduction problem and since

little additional complexity 1s introduced in the optimal analysts by adding

another control. this aspect is not dealt With in detail.

Section 2.4 considers non-interacting control. Since a major difficulty in

multivariable, feedback control design arises from the steady state and dynamic

interactions that occur between the various input and output variables. it is

usually desirable to reduce these interactions. If they can be reduced

sufficiently, single-loop control theories can be applied directly to each of the

non-interacting loops. The technique of perfect, non-interacting compensation

is attempted for this purpose. However since such compensation is in many

cases impractical or excessively complicated, other methods that only eliminate

steady state interactions are also considered. Finally, a new method, that uses a

relatively simple control structure to eliminate all steady state and dynamic

interactions, is considered. This procedure, called inner-loop decoupling, makes

use of extra available measurements through an inner-loop structure.

Several methods at the forefront of current multivariable frequency-

responsA analysis are also studied. Section 2.5 discusses the application of

Rosenbrock's (1962) inverse Nyquist array technique, which is an extension of

the classical Nyquist stability criterion. Section 2.6 analyzes the heat conduc-

tion system using the characteristic locus method. The analysis is based on the

latest refinements of the technique originally introduced by Belletrutti and

MacFarlane (1971). The current work provides a systematic approach for

designing a proportional-integral controller with the best compromise of system

stability, interaction, integrity, and accuracy.

Section 2.7 presents an overall analysis of the control system performance.

Using computer simulations of the responses of both the lumped model and the

actual system to step input changes, the effectiveness of the various control

designs are compared. This analysis leads to conclusions about the model

reduction and design techniques that can be extended to general, multivariable

feedback control.
 - 16-

2.2 PRELllliNARY ANALYSIS

Difficulties arise in the control system design of distributed parameter sys-

tems because of state variations in both time and space. The thrust of many of

the feedback control design techniques for distributed systems is to reduce the

system to a lumped one and then to take advantage of the many theories avail-

able for lumped parameter control design. However, problems arise in that all

of the analysis performed on the lumped system is dependent on the method

and accuracy of the reduction. Although considerable model reduction is neces-

sary to reduce computational complexities in the design procedures, excessive

or inaccurate reduction can lead to a system whose behavior is quite different

from that of the original process.

2.2.1 Yodel Reduction

To obtain the lumped parameter model for a system described by partial

differential equations, many efficient techniques are described in the literature.

In particular, much work has been published on various lumping strategies for

linear diffusion equations. A particularly useful means of treating both linear

and nonlinear partial differential equation systems is the method of weighted

residuals along with other pseudo-modal techniques, such as finite element

methods (Norrie and DeVries, 1973) or the use of spline functions (Finlayson,

1972). The method of weighted residuals is comprised of the following basic

techniques, depending on the choice of the weighting function (Prabhu and

McCausland, 1970; Ray, 1961):

a. Galerkin's Method (Lynn and Zahradnik, 1970; Newman and Sen,

1972: Prabhu and McCausland. 1970)

 - 19-

b. Method of Subdivisions

c. Method of Moments

d. Method of Collocation (Finlayson, 1972)

e. Least Squares Method.

Although these techniques are quite powerful, Mahapatra (1977) points out that

solutions using the method of weighted residuals often require considerable

effort to determine the set of orthogonal coordinate functions and a high·order

lumped model for accurate results. To eliminate these difficulties, spatial

discretization techniques (Leden, 1976; Mahapatra, 1977) are often quite useful

tor linear ditfus10n systems, since they retain the physical characterist1cs of the

system. However, they too often lead to high-order lumped models.

Thus to improve the accuracy and reduce the order of the model, it may be

best to use an exact reduction technique. Since the heat equation is governed

by a parabolic equation, exact lumping can be performed using a Laplace

transform in time or through a modal analysis. The latter, which is simply an

application of the separation of variable solution procedure, is quite attractive

for systems which can be made self-adjoint, since the technique leads directly to

the eigenvalues and eigenfunctions (modes) of the system. If the eigenvalues

are real, discrete and well spaced, the modal representation is a convenient

method to reduce the order of the system, since only the dominant modes need

be retained for design purposes.

Both the Laplace transform and modal analysis techniques were applied to

the one-dimensional heat conduction problem. Other techniques have been dis-

cussed in detail in the literature. Although the methods can be shown to lead to
equivalent results, the modal analysis leads directly to a lumped, state-space

representation. Disadvantages of the separation of variable technique were

cited by Prabhu and McCausland (1970), but the technique is well suited to the

linear diffusion problem because time and space variables are easily separated

and an analytic solution is possible.

Consider the system described by Equation (2.1-1), scaled so that y0 (z)=O.

Taking the laplace transform with respect to time gives:

sy(z s) = d2y(z,s) (2.2-1)

, dz2 ,

which has the solution

y(z,s) =A sinh ...JSz + cosh...JSz . (2.2-2)

After application of the boundary conditions

dy(O,s) -_ -u (s) dy(l.s) = u2(s) . (e.z-s)

1
dz dz

the solution in the Laplace domain is

y(z,s) = GJ'(z,s)u(s). (2.2-4)

where

(2.2-5)

T _ [ ] _ fl cosh( 1-z)-Vs cosh zv'S J. (2.2-6)

Gp- g 1(z,s) 'g2 (z,s) - v'Ssinhv'S ' v'ssinh..JS

Thus a distributed transfer function representation is obtained, from which a

simple feedback control strategy can be env{sioned (Figure 2.2-1). The closed-

loop distributed transfer function is

y{z,s) (2.2-7)
Yd(z,s)

However in general, measurements will only be available as discrete points. If

these points are Z~o ... , ZN. the appropriate block diagram structure is shown in
 - 21-

Figures 2.2-2 and 2.2-3 with

(2.2-8)

Note that Figure 2.2-3 is equivalent to Figure 2.2-1 with Gc(z,s) = Gc(s)A

Using contour integration to invert Equations (2.2-4)-(2.2-6), the time-

domain representation of the solution can be obtained. The inverse of g,;(z,s) is

the sum of the residues of estg1 (z,s). Each g,(z,s) has an infinite number of sim-

ple poles at Sn =-n2rr2 , n =1, 2, 3, ... and a pole at s 0 =0.0. Simple expansions of

the numerator and denominator of estg,(z,s) about s0 = 0.0 lead to a residue of

1. 0 at s 0 . The residues at Sn are obtained by Taylor expanding sinh ....; s about sn

and applying residue theory. This results in

g 1 (z.t) = 1.0+ 2:
-
2.0(-~)ncosnrr(l-z) e-n n
1!2t,
n=l
(2.2-9)
g2 (z,t) ::::: 1.0 + ~ 2.0( -1) ncos n1rz e-n2wl1.
n=l

Then using convolution theory. the t1me-domain behaVior is directly related to

the control action:

t
y(z,t) ::::: fo G(z.t--r)u(1)d1 G'l'(z,t) = [g 1 (z,t), g2 (z,t)] . (2.2-10)

Since the time-domain results are obtained as an infinite series of exponen-

tials with eigenvalues Xn = n 2 1r2 , the Laplace-domain behavior can also be

represented as an infinite series with y(z,s) being described by Equation (2.2-4)

along with the following:

GJ(z,s) = [.L+ f:
S n=l
2(-l)ncosnrr(1-z),
s-Xn
.L+ f:
s n=l
2(-l)ncosnrrz
s-Xn
l .( 2 .2 _11 )

Then if the series can be truncated after the first few terms without excessive

inaccuracy, normal multivariable design techniques can be applied directly.

 -22-

However since separation of variables for the one-dimensional heat conduc-

tion system leads to a self-adjoint operator With real. discrete eigenvalues,

modal decomposition is attractive for this system. The problem can be

redefined using the Dirac delta function:

By1~·t) =
02
~~~,t) + ~(z-o)ut(t) + o(z-1)ll2(t)
(2.2-12)
z = o-, z = 1+ ~~.t) = o .

It can be proven that this change is rigorous by integrating Equation (2.2-12)

across the infinitesimal intervals 1- < z < 1+and o- < z < o+. For example:

,.o+ B [ ~y)
;
,.o+ B
0
_ ¥t-dz = Jo- oz BZ]dz + fo-o-~-6(z-D)u (t)dz + Jo-
1
,.o+
c5(z-1)ll2(t)dz (2.2-13)

Thus

But ~"} = 0 at z = o- : therefore ~i = -u1( t) at z = o+. Thus formulation

(2.2-12) is equivalent to that described by Equations (2.1-1)- (2.1·3).

The space and time variables can then be separated by assuming a solution

of the form

..
y(z,t) ::: ~ an(t)~n(z)
n=O

t5(z-D)u 1(t) +o(z-1)u2(t) =I;

. bn(t)9'n(Z) . (2.2-14)
n=O

A!ter substituting into (2.2-12) and simplifying, the equations become

 -23 ~

n = 0, 1. 2, ... . (2.2-15)
z=O -'dscn(z)
-:;;.......;-=
dz
0' z =1

By choosing the separation constant as -An,

n=O, 1, 2, ... . (2.2-16)

Clearly (2.2-16a) is a self-adjoint differential equation that can be solved to yield

(2.2-17)

after application of the boundary conditions from (2.2-15). Because (2.2-16a) is

a homogeneous, self-adjoint differential equation With homogeneous boundary

conditions, the eigenfunctions, Equation (2.2-17), are orthogonal. It is con-

venient to choose the arbitrary constant An so as to make the eigenfunctions

orthonormal, i.e.,
1
fo sc~(z) dz =1.0 (2.2-18)

The appropriate choice of An leads to

1.0 n=O
Y'n(Z) = { -.n_ cos n1iz n= 1. 2 .... (2.2-19)

Then by application of the orthogonality of the eigenfunctions

1
e.n(t) =fo sPn(z)y(z,t) dz , (2.2-20)

or in particular for y0 (z) = 0, an(O) = 0. Similarly the coefficients bn(t) are given
by
 1
bn(t) = fo 9'n(z)[o(z-O)ut(t) +o(z-l)u2 (t)] dz
= 9'n(O)ul(t) +9'n(l)lll,l(t). (2.2-21)

Thus

n=O
n=1,2, ... (2.2-22)

Since the eigenvalues, 7\n =n21T2 , increase rapidly with increasing n, the sys-
tern can be accurately represented by the first few eigenfunctions, n = 0, 1. ... , N.

The process model can then be obtained as an Nth-order lumped state-space

representation with theN+ 1 states a0, ... , aN.

X(t) = Ax(t) + Bu(t) , y(z,t) = Cx(t) . (2.2-23)

where

1 1
'-'Z -..JZ

The resulting feedback control system can be drawn in block diagram form (Fig-

ure 2.2-4)

In theory, the control scheme of Figure 2.2-4 (and of Figure 2.2-1) requires

the complete temperature profile, y(z,t), of the system. However, Ray (1981)

points out several means of circumventing this problem. He suggests:

 -25-

L measuring y(7,,t) at many points i = 1. 2, ... , M and using an optimal

smoothing technique to approximate y(z,t),

ii. measuring y(Zt.t) at a few points and using a state estimator to esti-
mate y(z,t),

iii. or measuring y(zt.t) an N+l spatial points and letting

y(t) = [y(zt.t). y(z2.t), .... y(zNH.t)]

C= (2.2-24)

Then y(t) =Cx:(t) or x(t) =C 1

y(t) as long as z 1 .... , ZN+l are selected to

keep C nonsingular.
Actually it is possible to control the system by taking measurements at M points,

where M < N+l. For tbis case, the system should be controlled by using set

points on the outputs rather than on the states, since the N+ lth-order state
vector x:(t) cannot be obtained uniquely with M < N+l measurements.

Regardless of the technique for estimating y(z.t), the appropriate transfer

function representation can be obtained by converting to the laplace domain.

With Yo (z) = 0,

x(s) =(sl -A)-1 Bu(s), y(z,s) =Cx:(s) , (2.2-25)

or with measurements at M distinct points:

y(s) =Cx:(s) , C= (2.2-26)

 -26-

Thus the lumped parameter system process transfer function is given by

(2.2-27)

2.2.2 Controllability

The distributed system has been lumped through an N eigenfunction

decomposition. Before attempting to design a control strategy for the lumped

system, an analysis of system and output controllability is necessary.

Although the concept of controllability is formally defined in many refer-

ences (Brockett, 1970; Douglas, 197Z; Lee and Marcus, 1967; Ray, 1981), it is con-

venient to consider a system completely controllable if some control action

exists that will take the system from any given initial state to any specified final

state in finite time. The necessary and suffi.cient condition for complete control-

lability of the system described by (2.2-23) with N-1 states and two controls is

that the controllability matrix It:

(2.2-28)

1 0
' 1
..JZ -"2 -(_,.2~-JZ

has rank N-1. Since It: has full rank for all N, the system is completely controll-

able. Thus the two controls u 1 and u 2 are capable of influencing all of the states.

However, the lumped. analysis will be based on k measurements. Thus a

more important concept is that of output controllability, i.e., the controls must

be able to influence all k outputs. The output controllability matrix

 -27-

4' = [CBI CABI ... l CAN-lB] (2.2-29)

must have rank k for the system to be output controllable. It can easily be

shown that, for the heat conduction system, output controllability is assured if

no two measurements are taken at the same point.

It can be concluded that, for k distinct measurements, the lumped, one-

dimensional heat conduction system is completely controllable with heat fiux

control at z =0 and z =1. Thus multivariable lumped parameter control theory

can be applied to the state-space representation of the system. However, it

should be recalled that this analysis or controllability is dependent on the accu-

racy of the model reduction and lumping. Although the approximate lumped

parameter system has been shown to be completely controllable, the actual dis-

tributed system may indeed be only partially controllable. Additionally, care

must be taken in making conclusions from this type of analysis, since no con-

sideration of the physical constraints of the system have been made.

2.2.3 Root-Locus

The concept of root-locus analysis is basic to classical control system

design for single-input, single-output processes. The root-locus diagram is

advantageous since it describes the character of the response as the gain of the

controller is continuously changed, by allowing rapid determination of the roots

of the characteristic equation. Although scalar root-locus techniques are well-

known, the multivariable root-locus problem is relatively new. Kouvaritakis

(1978), Kouvaritakis and MacFarlane (1976ab), and Kouvaritakis and Shaked

( 1978) describe the technique and discuss the analysis of system zeros.

The objective of the root-locus method is to investigate the behavior of the

closed-loop characteristic frequencies when the feedback gain matrix has the
 -28-

form Gc = ki. For the modal-lumped representation of the heat conduction sys-

tem, the process transfer function was shown to be Gp =C(si-A)- B. 1

The

characteristic equation is then (Hsu and Chen, 1968)

(2.2-30)

where P0 (s) is the open-loop characteristic polynomial and Pc (s) is the closed-

loop characteristic polynomial. Then if we detine

si-A -B
z(s) = (2.2-31)
C Om,m

the n-m-d closed-loop characteristic frequencies 1 will tend toward the roots of

z(s) = 0 as k increases. These roots are the finite zeros of the process. For the

two-control. 3rd-order, heat conduction system, there is one finite zero and two

infinite zeros if CB has full rank. If CB has lost rank, all three zeros will be

infinite. This occurs for the following choices of measurement locations:

Z1 0.20 0.33 0.40 0.60 0.67 0.80

z2 0.60 0.67 0.80 0.20 0.33 0.40

Thus for any combination of outputs other than these, there will be one finite

zero given by z(s) = 0. The solution to this is

2
8
= (c-a)
41T
(2.2-32)
V2(ad-bc)+(a-c)

1. where n is the order of the lumped model, m is the number of inputs, and d is the rank deficiency
ofCB
 -29-

This root is finite for all z1 and Za other than those listed above, since the

denominator of Equation (2.2-32) is then nonzero.

The root-loci for the system are the loci of the roots of the characteristic

equation

+4k(ad-bc)]s +Brr 2 ~k(a-c)+Brr 4] (2.2-33)

as k varies from 0 to cc. Obviously, the poles (k =0) of the system are at s = 0,
-rr2 and -4rr2 independent of the measurements. As k approaches infinity, (2.2-

33) reduces to

~(ad-bc)s + (a-c)s + 4rr 2 (a-c) =0 (2.2-34)

which is equivalent to (2.2-32) above.

Figure 2.2-5 shows the root-loci for various measurement locations. Both

symmetric and unsymmetric cases were studied. For the symmetric cases, the

root-loci remain stable (in the left-half plane) at high gains for z 1 s: 0.33;

whereas, the loci become unstable at high gains lor 0.33 < z 1 < 0.50 . This is

expected due to the large lag time between the control action at z 1 = 0 and its

effect on the output. The symmetric cases with z1 and z2 reversed, i.e., z1 > 0.50

and z2 < 0.50 (not shown), lead to root-loci identical to those in Figure 2.2-5

except that the locus beginning at -1r 2 approaches +oo rather than -co. Thus

such a system is less stable. 2 Additionally from the root-loci analysis. the

2. This is also a result of the large lag times.

 -30-

responses for the (0.2,0.8) case are expected to be non-oscillatory; whereas, for

all other cases, oscillations are expected at moderate to high gains.

It can be concluded that, for symmetric measurements with z 1 < 0.33 or

for the unsymmetric case (0.4,0.8), system stability is insured even for high

gains. Thus a proportional controller may provide adequate control action.

Since the other cases lead to instability at high gains, more complicated control

schemes should be considered.

 - 31-

Yd(z,s) i'L E'(Z,S) u(s) y(z,s)

Gc(s) -... Gp(Z,s)
>(
-

Figure 2.2-1
Distributed Feedback Control Design

Process y(z,s)
Gp(Z,S)

Measuring
Device

Figure 2.2-2
Feedback Control Strategy with N Discrete Measurements
 -32-

Yd(z,s) Yd( s)+/0\ E( s) u(s) __ y(z,s)

A ~
Gc(s) Gp(Z,S)

Figure 2.2-3
Modified Feedback Control Strategy with N Discrete Measurements

Yd(z)+/0\ E(Z t)
Gc(z,t)
u(t) x= Ax +Bu y(z,t)
-
'<:1 ' y(z,t) =Cx( t)
-

Figure 2.2-4
Nth-Order Lumped Feedback Control Design
 -33-

I
.. 0 ..
g
...•

-.
0
0

-·
-uo. oo -no. oo -100. oo -10. oo ...o. oo -•o. oo -20. oo
.;

If
;
II
H.OO

0
0

...••
0
0
0
••

(a)

D
D
.,;
II
-
..
.,;
"'
.
-·
...---~-,,-.,----,---i--..-----.
-2110. 00 ..,.o. 00 -200. 0 -160.00
.........--....--~-+-
-120.00 .eo. 00
.,;
...
....r--"........:.,•
40.00

,..
0

.,;

.""
.,;
:!:
•

(b)
Figure 2.2-5
Root-Loci Diagrams for Various Outputs
a. (0.2,0.8) b. (0.3,0.7) c. (0.4,0.6) d. (0.4,0.8)

x- Poles 0- Zeros
 -34-

8
! ..
g

...
-1to. oo -JJo. oo -eo. oo 011 2411. 110 J:lO. 011
·-
tOO. 110

(c)

....
D
D

II D

"'.."'
!:

..."'
-·
-uo. oo -120. oo -too. oo -eo. oo -60. oo -to. oo
..
•20..00
'I'

."'-..
0

."'.
..,
I

(d)
Figure 2.2-5 Continued
 2.3 TlME-DOM.AlN ANALYSIS

The time-domain approach for control system analysis utilizes the

differential or difference equations directly rather than using transfer functions,

as in the frequency-domain analysis. Although the time-domain technique is

actually older than frequency response techniques, its development was slow

due to the difficulty of making calculations in the differential domain. :rhe

emergence of digital computation as a widely accepted tool led to a resurgence

of interest in time-domain analysis. In particular, the techniques of state esti-

mation, optimal control, modal control, and adaptive control rose to the fore-

front of research. Although these methods have a strong theoretical basis, they

only became practical with the development or small and reliable digital com-

puters: capable of high-speed information processing. Consideration of ideas for

which frequency-domain techniques were inappropriate, such as simultaneous

control of several interacting variables. and the application of different types of

controller objectives, such as the minimization of energy consumption, became

practical.

Of the many time-domain procedures. optimal control and modal control

are the most common techniques and thus have been studied extensively for

both lumped and distributed parameter systems. The one-dimensional diffusion

equation has often been used to illustrate the application of these methods.

Due to the many studies of these theories and on their application to heat con-

duction systems, only a cursory examination of the techniques will be

presented. Additionally. other methods such as adaptive control and state esti-

mation will not be considered in this theoretical analysis of the heat conduction

process. although they may be quite useful in practical applications.

 -36 ~

2.3.1 Optimal CGntrol

Optimal control methods can be divided into two schemes-open- and

closed-loop. When an excellent mathematical model of the system is available in

terms of differential equations, open-loop control schemes can be useful for

start-up, shut down, and other transient conditions. However in practice, most

models contain some error; therefore, closed-loop schemes are often necessary

for satisfactory controller performance. since they involve feedback of process

measurements. Regardless of whether open or closed~loop control is to be used,

the technique involves the selection of an index which measures the perfor-

mance of the system, from which the optimal control strategy is selected as that

which minimizes this index. A major difficulty in the design of an optimal con-

trol system is the establishment of the criteria for optimality. The optimal con-

trol procedure is in contrast to the other techniques that try to obtain satisfac-

tory responses in terms of offset, gain margin or decay ratios, since once the cri-

teria is selected a unique solution is obtained.

As previously mentioned, much work has been published concerning

optimal control of parabolic systems such as that described by the heat equa-

tion. McCausland (1970), Prabhu and McCausland (1970), and Mahapatra (1977)

studied time-optimal control of the linear diffusion process. For such control,

the objective is to force the system to reach the desired design conditions in

minimum time. Others (Betts and Citron, 1972; Sakawa, 1964: Sheirah and

Hamza, 1974) treated the problem of optimal control of the heat conduction

problem by minimizing the deviation of the temperature distribution from the

desired distribution throughout time. Additionally much literature is available

on the general problem of optimal control of distributed parameter systems.

An important special case of the optimal control problem is linear~

 -37-

quadratic control, which leads to an optimal, state-feedback control law.

Numerous papers have been written on this problem, including several on its

application to parabolic systems (Ahmed and Teo, 1981; Matsumoto and Ito,

1970; Wang, 1975). This technique uses a quadratic penalty function to control a

system at a set point without excessive control action and not exceeding accept-

able levels of state. The method is readily applicable to either a lumped model

of the system or to the original distributed model.

For the heat conduction system, the lumped parameter model is described
by the linear differential equation

i=Ax+Bu, y=Cx 1 (2.3-1)

with A. B, i:ind C defined by Equations (2.2-23) and (2.2-26). The objective of this

technique is to obtain the feedback law which minimizes the performance index

(2.3-2)

where Sp, F(t), and E(t) are symmetric, positive definite weighting matrices

which describe the relative importance of reaching the desired set point Xo. = 0,
using small levels of the state and using little control action. If the desired set

point is nonzero, then deviation variables can be used to convert the problem to

the above form. Thus let Xo. and lld be the desired steady state values. Then let-
ting

u' = u -Uci (2.3-3)

and recognizing that at steady state Aid + Blld. = 0, Equation (2.3-1) becomes

i.' =AX+ Bu' (2.3-4)

with the performance index:

 -sa-

Then from quadratic feedback control theory (Bryson and Ho, 1969; Ray, 1981),

the feedback control law is

u(t) = ~ - K(t)(x- X<~.) (2.3-6)

where K(t) is given by

K(t) = E-lJil'S(t) (2.3-7)

and S(t) is found by solving the Ricatti equation backward from tF:

s(t) =-SA- ATs + sm:-t]jl's- F. S(tp) =Sp. (2.3-8)

Thus a proportional feedback controller with time-varying gain has been

designed to control the system while minimizing the index given by Equation

(2.3-5). This control structure is quite useful because the lime-varying gain K(t)

can be determined otT-line since it does not depend on x(t) or u(t). Then if tp """

oo and A, B, F, and E are constant, S(t) becomes constant. It is the solution to

sm:- 1J3l'S- SA- ATS- F = 0 (2.3-9)

In this case, the controller is simply a constant gain proportional controller.

The linear-quadratic problem can also be applied to the distributed param-

eter system described by Equations (2.1-1) - (2.1-3). The objective is then to

minimize the index

(2.3-10)

where y· and u' are deviation variables With Yd(z) being the desired temperature
 -39-

profile. The derivation of the optimal control law is carried out using the pro-

cedure described by Ray (1981). The results are:

=-Eo-1 fo
1
u 1 (t) S(O,s,t)y(s,t) ds
(2.3-11)

where S(r,s,t) can be computed ofHine from

St.(r,s,t) = -S88 - Srr + S(r,O,t)E0 1S(O,s,t)

+ S(r.l.t)E} 1S(l.s,t)- Fo(r-s) (2.3-12)

with the boundary conditions

Sa(r.l.t) =S.(r,O,t) =Sr(O,s,t) =Sr(l,s,t) =0 (2.3-13)

and terminal condition

S(r,s,tr) = Sr(r,s) = S,O(r-s) (2.3-14)

Thus linear-quadratic optimal control can readily be applied to the heat

conduction system to obtain the feedback control law using either the lumped

or distributed parameter model. However several problems exist with the

optimal control technique. MacFarlane (1972) points out that optimal controll-

ers provide gain margins far in excess of those required for stability, are often

difficult to tune on-line and may be of low integrity to transducer failures. Addi-

tionally there are several other major concerns:

i. Optimal control design requires an accurate model of the system. This can

lead to difficulties in the lumped parameter design due to errors intro-

duced by model reduction. Even in the distributed parameter design,

model inaccuracies could be quite large due to heat losses or inefficiencies

in the controls.
 ~ 40-

ii. Optimal control design requires all of the system states to be JiCcessible.

Thus for the lumped model. the technique is restricted to the case where
the number of measurements equals the order of the model and the meas-

urement matrix is nonsingular. Only then are the states accessible from
the measurements:

(2.3-15)

For the distributed parameter design. the entire temperature profile y(z.t)

is needed. Considerable effort has been directed at overcoming this

difficulty by using observers or Kalman-Bucy f:Uters to recover the inacces-

sible states. Much literature has been published on combining such tech-

niques with optimal linear quadratic control.

iii. Optimal control design requires a selection of the weighting matrices Sr. F,
and E. Unfortunately, in many chemical engineering applications. the

choice of the weighting matrices may be quite difficult. For the heat con-

duction system. no easy criterion is available for selecting the weights.

Consequently. much of the literature dealing with optimal control of

diffusion systems considers simple minimization of the time needed to

reach the desired state or the deviation of the system from the desired

state.

2.3.2 llodal Control

Early work in modal control was dominated by Rosenbrock ( 1962) and

Gould and Murray-Lasso (1966), with many others (Bradshaw and Porter, 1972;

Davison, 1970: Ellis and White, 1965abc; Fisher and Denn, 1978: Porter and

Bradshaw, 1972) extending the basic theories. Much work has also been pub-

lished on the application of modal techniques to various systems ranging from

linear diffusion problems (Balas, 1979; McGlothin, 1974; Porter and Bradshaw,
 -41 -

1972; Wang, 1972) to large chemical plants (Davison and Chadha, 1972).

Although many texts discuss the concepts of modal analysis, Gould ( 1969) pro-

vides a detailed discussion of the use of the method for distributed systems and

for lumped systems with an arbitrary number at states, controls, and measure-

ments. This latter situation is of great interest for the heat conduction prob-

lem, since the order of the model may be much larger than the number of con-

trols and measurements.

Modal analysis is based on the postulate that the transient behavior of a

process is primarily governed by the modes associated with the smallest eigen-

values and that the response of the system can be improved by using a control
design to increase these eigenvalues. Additionally, the method suggests that it is

possible to approximate a complicated, high-order system by a lower-order sys-

tem whose slow modes are the same as the original system. This technique was

used in the modal lumping of the previous section to obtain the Nth-order

lumped model for the heat conduction system. However several references

(Douglas, 1972; Gould, 1969; Ray, 1981) point out that disturbances affect the
different modes differently; therefore, the reduced model may not be satisfac-

tory if the disturbances have their greatest effect on the neglected faster modes.

Additionally, although Rosenbrock's ( 1962) approach implied the possibility of

altering each eigenvalue separately without limit, this is often impractical owing

to limitations on the number of controls and measurements and their locations.

The ideal case for lumped modal analysis is when the number of controls

and number of measurements equal the order of the model. For the system

described in Equation (2.2-23):

.i: =Ax+ Bu, y=Cx (2.3-16)

the modal technique calls for the selection of C = L, where Lis the matrix of left
 -42-

eigenvectors of matrix A. The controller is then designed as a simP.}e propor-

tional controller with

(2.3-17)

where R is the matrix of right eigenvectors of A and K is a diagonal proportional

gain matrix with diagonal elements kt. With these selections.

:i = R(A - K)Lx. .t = (A - .K)y (2.3-18)

where A is a diagonal matrix of the eigenvalues. Thus the outputs have no

interaction, and the eigenvalues have been shifted by k,. Since A is a diagonal

matrix for the heat conduction system. A =A and the matrices R and L can be
simply taken as identity matrices. Thus the technique calls for the selection of

the measurement matrix as an identity matrix. Therefore, an appropriate com-

bination of the temperatures should be used so that the system states are actu-

ally measured. 1

However in most practical systems, the number of states will exceed the

number of controls. Consider the heat conductton process With two controls

and N states and measurements. Gould (1969) shows that, if B and C can be

chosen arbitrarily, the two lowest eigenvalues can be made as large and negative

as desired while leaving the others and all the eigenvectors unchanged. As shown

above, the restrictions on C can be accommodated by measuring a combination

of the temperatures. However for the heat conduction system, B is fixed due to

the a. priari selection of the controls. By slight modification of the analysis,

modal control is still applicable. Consider adding both a compensator G.c to

adjust for the fixed B and a diagonal proportional control matrix K:

1. Thus the compensator c-t should be included after the process and set points or after the
difference junction
 -43-

u= -Gc,Ky with set point Yd = 0. (2.3-19)

Note that previously when B had dimension N. 2 Gc was selected as Gc = ~ R.

1

However in this case, B can not be inverted. Thus

i = (A - BGcK)x (2.3-20)

with A:::: A and C selected as C = L. 3 Then the choice of

_l_ 1
2 -2v2 0 ... 0
Gc= (2.3-21)
1
_l_
2 - 2..J2 0 ... 0

leads to

-kl 0 0
0 -k2 0
-"'Zkt 0 0
0 -..J2~ 0
BGcK = -"'Zkl 0 0
(2.3-22)

Thus the first two eigenvalues can be altered by arbitrary selection ot k 1 and k 2

without affecting the higher N-2 eigenvalues; however, the first two eigenvectors

have also been altered slightly. Thus a disturbance in either of the first two

modes will cause a disturbance in the higher modes. Since they decay rapidly.

little difficulty should result. The appropriate block diagram is shown in

Figure 2. 3-1.

Gould ( 1969) also presents a detailed discussion of the situation where the

number of measurements is less than the dimension of the process. This would

2. B Wlill invertible.
3. Note that Lis taken as the identity matrix.
 -44-

obviously lead to difficulties in the modal analysis since the system state cannot

be obtained from the measurements. An appropriate control scheme can be

designed but will lead to considerable interaction between the control loops.

Because of the complexity of the resulting control scheme and the ease of

obtaining temperature measurements for a heat conduction process, it is

recommended that sufficient measurements be taken so that the state vector

can be constructed. Balas ( 1979) discusses the introduction of a Luenberger

observer for this purpose.

Additionally, modal control can be applied directly to the distributed

parameter system. Gould and Murray-Lasso ( 1966) and Gould ( 1969) present a

detailed discussion of modal control for linear, distributed systems. However,

using these techniques and considering the limitations imposed by taking only a

finite number of discrete measurements and by manipulating only two controls,

the analysis becomes equivalent to using modal lumping and applying lumped

modal analysis.

Although modal analysis leads to exact placing of the poles, the procedure

leads to a simple proportional controller and can not give any guidance as to

the selection of additional control action. More importantly, the technique uses

no information concerning the zeros of the transfer functions. Problems can

easily arise if the zeros of the closed-loop transmittances move into the right-

half plane. Furthermore, the method gives the best results only if all the states

of the system are accessible. Although this does not cause any difficulty for the

heat conduction process, it is not practical for most systems.

 -45-

u Process y
..!L.
-*
c-1 ltd+
K Gc G.,

X c-1

-•
+ u Process y
c-' KGc Gp

Figure 2.3-1
Lumped Modal Feedback Control Design
 -46-

2.4 NON-INTERACTING CONTROL

A major difficulty in multivariable, feedback control design arises from the

steady state and dynamic interactions which occur between the various input

and output variables. In most systems, it is desired that one specific output

y;.(s) responds to input U;.(s). while all other outputs remain unchanged. The

term interaction can then be used to refer to the effects that a particular input

U;.(s) has on the output Yi(s), j ~ i. Thus low interaction is usually desirable. In

fact if it is possible to eliminate all the coupling between variables. and if the

number of controllable inputs and outputs is equal, the multivariable system

can be treated as a combination of single-loop systems, and classical techniques

can be used to tune each loop. Unfortunately, most multivariable systems have

considerable interaction; thus, several techniques have been devised to elim-

inate or at least reduce the interaction to an acceptable level.

For the heat conduction process, analysis of the system transfer function

Gp ;;;; C(s1-A)- 1B shows that interaction is high and cannot be eliminated by sim-
ple selection of the measurement locations. Thus compensators that reduce or

eliminate interaction may be useful. Several design techniques including per-

fect non-interacting compensation, steady state decoupling. and set point com-

pensation are available for this purpose. The application of these methods to

the heat conduction problem is discussed. Additionally, a new technique is

presented that eliminates interaction by using extra measurements within an

inner loop, leading to a relatively simple control strategy.

2.4.1 Perfect Dynamic Compenaation

Consider the third-order model for the heat conduction system with two

controls and measurements. The Laplace-domain representation of the system

 -47-

is

y= Gpu , (2.4-1)

with

(2.4-2)

B= [~ -:J (2.4-3)

(2.4-4)

The process transfer function Gp is

1 ...JZc12 ...JZc1s ...JZc12 ~c1s

-+ 2 + +
s s + rr s +4rr 2 s s + 1'1'2 s +41'1'2
Gp = 1 ~Cz;:: ~c23 ~c22 ~Cza
(2.4-5)
-+ + +
s s + rr 2 s + 4rr2 s s + 1'1'2 s +41'1' 2

As mentioned earlier, the off-diagonal terms are significant and cannot be elim-

inated with the selection of the measurement locations z 1 and z2 .

Consider the feedback control system (Figure 2.4-1) consisting of single-

loop controllers, represented by a diagonal transfer function matrix Gc. and a

non-interaction compensator Gt proceeding the process. The closed-loop

response for this scheme is

(2.4-6)

The compensator Gt should be designed to eliminate as much interaction as pos-

sible. Ideally ·~"" and to drive Gc1
Gt should be selected to make Gc1diagonal -+ 1 for

s = 0 (t -+ ""). Note that such analysis is only applicable to systems with square

transfer function matrices. If there are more controls than outputs, then a sub-

set of controls may be chosen for decoupling, while if there are more outputs
 - 4-B-

than controls, only partial decoupling is possible (Ray, 1981).

It Gc is diagonaL a sufficient criterion for GcJ to be diagonal e.nd GcJ(O) ... I as

the controller gains increase is to select

~=Gp 1 diag Gp(s) , (2.4-7)

where diag Gp(s) is a diagonal matrix of the diagonal elements of Gp(s). If this

decoupling is performed perfectly, the closed-loop response will obey

(2.4-8)

where di(s) are the disturbances and g~(s), Sc:«(s), and gp«(s) are the elements

of the disturbance, controller, and process transfer function matrices, respec-

tively. This selection of Gx leads to a total decoupling for set point changes and,

even though each disturbance can intluence all the outputs, its effect on output

Yi is damped by a single controller gcu(s) (Ray, 1981).

Although this method is simple and seemingly eliminates all the complica-

lions inherent in multivariable, feedback control design, it has several major

disadvantages. Perfect compensation requires a fr~rfect transfer function

model. Though the model is well-known for the heat conduction process,

approximations such as finite lumping can lead to badly behaving or even

unstable control schemes. Furthermore, MacFarlane (1972) and Ray (1981)

point out that another potential disadvantage is that a great deal of control

flexibility is used up in making GpGtGc diagonal. often by sacrificing closed-loop

dynamic performance. In many cases some interaction may actually improve

dynamic performance. Finally. the technique possibly leads to an unnecessarily

complicated compensator and fails if the determinant of Gp has right-half plane

zeros or if the transfer functions are not square. Thus other techniques may

lead to a simpler control scheme which is easier to tune on-line and less
 -49-

sensitive to model inaccuracies.

2.4.2 Steady State Decoupling

Although perfect compensation may be impossible or merely impractical.

steady state decoupling is usually quite useful and can be ;;Implemented before

applying other design techniques such as inverse Nyquist array or the charac-

teristic locus method. Steady state decoupling uses a compensator

Gr =lim [Gp1(s) diag Gp(s)] (2.4-9)

.-.o

to eliminate steady state interaction. However, it cannot improve dynamic

behavior. Unfortunately, due to the perfect symmetry of the heat conduction

system, even this method leads to difficulty. For the third-order heat conduc-

tion system, the appropriate compensator is singular:

M(c12-cz2)
Gx = -rr2 (2.4-10)
2-.t!(c12-cz2)

Introduction of this compensator into the control loop leads to dependent con-

trol action and is therefore not advisable, since it will lead to excessive dynamic

interaction and poor closed-loop performance.

2.4.3 Set Point Compensation

Steady state interaction arises because a change in the set point of one

controlled variable affects all the system outputs. A means of eliminating steady

state interaction could be to simultaneously alter all the other set points in

such a way as to cancel the effect of the original change. This idea is the basis

for set point compensation, which is used to eliminate or minimize steady state
 -50-

offset due to set point adjustments.

Consider the control scheme of Figure 2.4-2, where Yd is the actual set

point desired, Gc is a diagonal controller matrix, and G. is the set point compen-

sation matrix. The overall closed-loop transfer function is

(2.4-11)

The objective is to select G8 so that Gc1 is diagonal at stead~ state. Using the

theory of laplace transforms,

where

Gas= limGp
s-tO
(2.4-13)

For the heat conduction system,

Note that c 12 # c 22 if z 1 # z2 . Although this compensator is realizable and does

not lead to dependent control, it cannot improve the response to a disturbance

in the system since it does not appear in the feedback loop.

2.4.4 Inner-Loop Decoupling

Several recent studies (Foss et al., 1980; Kouvaritakis et al., 1979) discuss

squaring down extra measurements in an inner loop to obtain a system that

shows better control properties from the viewpoint of the outer loop. Usually

the inner loop is used to adjust the poles of the system, and then other fre-

quency response techniques are used to design the outer-loop control for the
 -51 -

improved process. However, the method presented here uses the additional

degrees of freedom inherent in adding an inner loop to eliminate or minimize

the interaction, with only simple proportional gain compensators and without

severely limiting the design freedom available for the outer loop.

Consider the process sht wn in Figure 2.4-3, with an outer loop consisting of

two outputs and two controls. This is identical to the system used by the previ-

ous methods. Using the third-order model for the process, a third independent

measurement can be made. The three measurements are denoted by YJI. The

following relationships describe the control scheme.

(2.4-15a)

[1 0 OJ
L= 0 1 0 · (2.4-15b)

Let Up be the transfer function from w toy, or essentially the transfer function
of the process seen by the outer loop. The objective of the technique is to select

the elements of F to reduce or eliminate interactions, i.e., select F so that Gp has

small off-diagonal elements.

The overall closed-loop transfer function for the inner loop is

(2.4-16)

If F is taken as

(2.4-17)

then

__ [g1(z1,Z2,f21•f22·f23) g2(zl,Z2,fu.ft2•fts)]
Gp- !?;a(zl.z2.f:.u.f22.f2s) g.(zl.z2.fu.ft2•fts) (2.4-18)

To eliminate interaction, there are eight degrees of freedom available--the

 -52-

locations of the two major measurements and the six elements of the squaring

down compensator F. The location of the final measurement can be left to the

discretion of the designer since it will only aff!"lct the components of F once F is
~

specified. Obviously, z5 must be selected different from z1 and z2 so that C is

nonsingular. For the third-order model of the heat conduction system, the

measurement locations and the elements of F should be selected so that the

(1,2) and (2,1) terms of Gp are identically zero. Since the model is third order, it

takes three degrees of freedom to make each term identically zero. Thus of the

eight available degrees of freedom, six are used to eliminate interaction, and the

remaining two can be used to improve the performance (move the poles) of each

non-interacting loop ..!l'urther analysis shows that

To eliminate term (1,2):

To eliminate term (2,1):

The additional relations:

c 12 ; ~cos rrz 1 c 1s ;;;;; ~cos 2rrz 1

c22 = ~cos rrz 2 c 23 = ...i'Zcos 2rrz2 (2.4-19)
 -53-

lead to the requirements that z 1 =0.2 or 0.6 and ~ =0.4 or 0.8. Note that the

elements of F are all simply constants. Thus the following situatio,"..s are allow-
able for decoupling of the loops

zl z2
0.2 0.4 Case I: c 12 =-c2s. cul=c22
0.6 0.8 (2.4-20)
0.6 0.4 Case II: c12=-c22· c1s=c2s
0.2 0.8

These rules lead to considerable simplification in the transfer functions:

0
(2.4-21)
0

0
(2.4-22)
0

Thus, not only has interaction been eliminated by using simple gain in an inner

loop, the new process is quite simple, and two degrees of freedom, along with the

location of z3 , remain at the disposal of the designer. The selection of fu and f 21

can easily be used to move the poles of each decoupled loop.

The tlnal design then involves a squaring down compensator l' which

together with the measurement matrix C leads to the desired F':

1T2 "'2 2 ""2rr2-ct2f u

fu c12fu-- c12fu- 1T + ~
2~ C12
l'= (2.4-23)
,.e 2 ~11' +c22f21
2

f21 c22r21 + 2""2 -c22f2l-..J21T

..J.Zc22
 -54-

The fact that F is a constant matrix is of great importance. If more degrees of

freedom are desired than the two available with this technique, F can be

designed as a function of s. For instance, term (1,2) could also be eliminated by

letting

with no restriction on f 12 or z1 . However, this leads to an unnecessarily compli-

cated feedback compensator for the third-order heat conduction system.

Inner-loop decoupling is an improvement over the perfect, non-interacting

control scheme for the heat conductions system. Although all of the techniques

can be easily applied, the inner-loop decoupling strategy may lead to a very sim-

ple control structure for processes where extra measurements can readily be

obtained while leaving considerable design freedom available for tuning the

dynamic behavior of the overall system. The only restriction on the extra meas-

urements is that all outputs must be linearly independent. Additionally, the

method still sutiers from the apparent necessity of an accurate process model.

However, as we will show in Section 2. 7, although the design is conducted on the

approximate, third-order model for the heat conduction process, interaction is

still nearly eliminated for the actual system. In fact, simulations also show low

sensitivity to the measurement locations. Thus the inner-loop scheme may be

superior to conventional non-interacting control methods.

Further analysis shows that an inner loop with constant feedback gains

cannot be designed to decouple the heat conduction system in the absence of

 -55-

extra measurements. However for such a case, a feedback compensator that is

a function of s can decouple the system but is of the same complicated compli-

cated form as lhal designed by the perfect, non-interacting control scheme.

Thus without extra measurements, inner-loop decoupling has no advantage over

conventional techniques.

Inner-loop decoupling still suffers from some of the inherent problems

associated with the field of non-interacting control. A non-interacting system

may no~ be desirable for chemical processes, since the major objective in such

processes is often the reduction of the etrects of disturbances on the system.

Although non-interacting analyses can lead to simplifications that allow comple-

tion of the design using single-loop approaches, there is no intrinsic reason for

non-interaction to improve control. Actually, exploitation of the interaction

among the variables may be usefuL Furthermore, the determinant of the pro-

cess transfer function matrix for multivariable chemical processes often has

right half plane zeros, which lead to poor or unstable control performance with

the non-interacting techniques.

 -56-

d Disturbance 1--
Gd

yd "tt< . Controller Interaction

Compensator
u Process
Gp +Jf
.,..+ y
'<?'- Gc
Gt

Figure 2. 4-1
Perfect Non-Interacting Control Strategy

y;, Set-Point Yd +J(:>. « Controller Process y

Compensation '<~ Gc r--- Gp

Figure 2.4-2
Set Point Compensation
 -57-

.,
-•
yd + Controller w +.o("' u Process YM -
Gc '<; - Gp L

YF
F

Figure 2.4-3
Inner-Loop Decoupling
 -58-

2.5 INVERSE NYQUIST ARRAY

Since classical frequency methods have proven extremely useful for design-

ing single-loop control systems, much work has been devoted to extending these

techniques to multivariable systems. The Nyquist diagram, which is a polar plot

of the information presented in a Bode diagram, is an excellent classical method

for determining system stability criteria for single-loop processes, since both

the magnitude and phase information appear in a single curve. Using the
Nyquist representation, it is fairly simple to determine the process stability

characteristics and the closed-loop dynamics from the graph for the open-loop

process.

The inverse Nyquist array (I.N.A.) technique, introduced by Rosenbrock

(1969), is a useful extension of scalar Nyquist array methods to the design of

multivariable control systems. It allows for considerable flexibility, is insensitive

to model inaccuracies, and reduces to traditional methods in the single-loop

case. Also, the technique is able to handle models specified only in terms of a

limited amount of directly obtained experimental data (MacFarlane, 1972) and

can be easily incorporated into a computer-aided design package.

Assuming that the open-loop transfer function Q(s):::: Gp(s)Gc(s) is nonsingu-

lar, which is necessary for output controllability, the inverse of the closed-loop

transfer function is

R- 1 = R= I+ Q. (2.5-1)

The notation Q :: Q- 1 has been used for convenience, since in general

~ A
qiJ 1 (s) ¢ <t-u(s), where qv and qij denote the elements of matrices Q and Q,

respectively. Thus the elements of Rcan easily be found from the elements of
 -59-

~·(•) = ~(s) + 1.0

__
rij(s)
_
=qij{s)
l i, j = 1, 2, ... , m,

where m is the number of controls and outputs. Furthermore if the kth feed-

back loop is opened, fA:.t(s) :::: ~(s). Then the I.N.A. is the set of m 2 diagrams

representing the loci in the complex plane corresponding to the elements of

Q{jc..>). In terms of frequency-response plots, the I.N.A. allows easy determina-

tion of the elements of R(jc.>), whether the feedback loops are open or closed.

The basis of the LN.A. design technique is Rosenbrock's stability

theorem (Rosenbrock, 1969):

Let D be a contour in the complex plane consisting of the imaginary axis

from -jex to +jex and a semicircle of radius ex in the right-half plane. where

ex is sufficiently large to insure that all finite poles and zeros of IQ I , IR I,

qij, qiJ', rij, and fiJ' lying in the open right-half plane are within D and

those on the imaginary axis lie on D. Then a feedback system will be

closed-loop stable if the system is open-loop stable and if Ge(s) is

designed such that

i) the inverse Nyquist mapping, ri. of each diagonal element q-n{jc.>)

of the I.N.A. for Q{jc.>) encircles the point (-1.0,0.0) the same

number of times in the same direction as it encircles the origin.

ii) fori= 1, 2, ... , m and for all son D

lcta(s)l > ~ lq'LJ'(s)l (2.5-3)

i='tJ"'
iii) for each loop j which is closed and for all s on D

(2.5-4)

The conditions (ii) and (iii) lead to diagonal dominance and thus insure that
interactions are sufficiently small to allow stability (condition (i)) to be deduced

from the diagonal elements of qj.,; alone.

For the lumped model of the two-control. heat conduction process, the cri-

teria for diagonal dominance are that

lq11 1> lq12 1 lqll + 11 > lct,21

IQ22 1> IQ2 11 1422+11> ltJ2ll
for all s on D. These conditions can be represented graphically (MacFarlane,

1972) as in Figure 2.5-1. Thus diagonal dominance is insured if the origin and

the point (-1.0,0.0) are not within or on any of the circles. Closed-loop stability

can then be checked from the two diagonal entries of Q(jc.>). The system is stable

if rl and r2 satisfy the encirclement criteria.

The I.N.A. design method involves adding controllers to make the system's

open-loop transfer functions diagonally dominant. The remainder of the design

is completed on the basis of a set of individual single loops using conventional

single-loop inverse Nyquist techniques. The I.N.A. method thus leads to a stable

system which has high integrity and low interaction, when diagonal dominance is

imposed.

The third-order model of the heat conduction process was analyzed using

the I.N .A. procedure with various measurement locations (Table 2.5-1). Both the

symmetric (Group I) and the unsymmetric (Group II) cases were considered with

z1 < 0.50. For the situation where z 1 > 0.50, a permutation matrix should be
used to interchange the inputs and outputs.

Due to the symmetry of the system for Group I. only the q11 (s) and q12 (s)
elements of the I.N.A. are needed since qz2 (s) and Cf2 1(s) are ident~cal. Addition-

ally to aid in the design procedure, logarithmic plots of Ict, 1 1. l1 + q11 1, and lq12 1
versus the frequency were also used. These are easier to use to check for
 - 61-

Group I GroupD

ZJ z2 ZJ z2

0.10 0.90 0.40 0.70

0.20 0.80 0.30 0.80
0.25 0.75 0.40 0.80
0.30 0.70
0.33 0.67
0.40 0.60

Table 2.5-1
Measurement Locations

diagonal dominance than the type shown in Figure 2.5-1. Although all of the
cases in Group I were analyzed, only two need to be considered in detail since
the others are similar. Figures 2.5-2 and 2.5-3 show the I.N.A. plots for measure-
ment locations (0.3,0.7) and (0.4,0.6). These cases were considered since the
root-locus analysis showed that the (0.3,0.7) case is stable while the (0.4,0.6)
case becomes unstable for gains above about 33.0. The I.N.A. analysis for the
cases in Group I verify the root-locus results that stability is insured at high
gains !or the symmetric case if :t: 1 < 0.33. Note that the I.N.A plots are only
drawn for s = c.>j with CJ = 0 ... IX) , since s = -c.>j simply gives the reftection.

For the cases with z1 < 0.33 (z 1 = 0.3 in particular), the conditions for diag-
onal dominance are satisfied for all s on D except at s = 0. At this point.

q11 = q12 = ci2 1 = ci22 . Further analysis shows that the problem that arises at
s = 0 is due to the pole which is at the origin in our process model. With these
problems at s = 0 eliminated, the cases tor z1 < 0.33 would be diagonally dom-

inant, and the system would be stable since the encirclement criterion is
satisfied.

One method of analysis is suggested by MacFarlane and Postlethwaite

(1977). They show that when a pole occurs on the imaginary axis, the modified
 -62-

Nyquist contour, D', shown in Figure 2.5-4 should be used. Then the conditions
for diagonal dominance must be examined in the regions:

a) s = c.>j
b) s =eel'«' where R.-. ao and t ... 0.

c) s =Rei'~'

Mathematical calculations for these regions indicate that the conditions for
diagonal dominance are satisfied for z 1 < 0.33 with z2 = 1.0 - z1 . Then since the
encirclement criterion is also satisfied, stability is assured. The I.N.A. technique
then concludes that elimination of interaction is unnecessary since arbitrary
high gains can be applied to each of the two principal loops Without instability
and Will lead to a system with little interaction.

Figure 2.5-3 shows that the (0.4,0.6) case does not satisfy conditions (li) and
(iii) of Rosenbrock's theorem (1962), since lq12 1is larger than lq11 1and lqi 1 + 11 for
some values of c.>. Although condition (i) seems to show closed-loop instability
for this process, no conclusion can be drawn since the stabllity criterion i.s a
sufficient condition and only becomes necessary if (ii) and (iii) are satisfied.

The problem then is to· design a compensator, Gc(s), so that the system's
open-loop transfer function Q(s) is diagonally dominant when 0.33 < z 1 < 0.50.
Using a general 2 x 2 compensator, the restrictions on the elements of Gc(s) can

be derived. Specifically consider the case (0.4,0.6). Using the modified Nyquist
contour, it can be shown that s = 0 presents no difficulty. Figure 2.5-3 then
shows that conditions (ii) and (iii) are satisfied at low CJ but tail at high CJ. Using
the restrictions on the elements of Gc(s), it can be shown that, owing to the sym-
metry of the system, it is not possible to design a nonsingular constant matriX
Gc(s) to make the system diagonally dominant for all s on the modified Nyquist
 -63-

contour. However, the analysis also shows that a compensator with the following
properties will lead to diagonal dominance:

As t:.)-+ 0 Gc(s) -+ [~ ~].

(2.15-6)

Gc(s) -+ [~ ~].
After many controllers were tried, the best results were obtained with

5.0 s
s+5.0 s+5.0
Gc(s) = s 5.0 (2.5-7)
s+5.0 s+5.0

Figure 2.5·5 shows the l.N.A. plots with this compensator. The controller has
indeed made the system diagonally dominant for all s on D'; however, the enctr~

clement criterion shows that the system becomes unstable for moderate gains.
This was verified using root-locus and characteristic locus analyses for the sys-
tern with this compensator. Attempts at eliminating the instability were futile,

and further analysis showed that the stability problems arise from an unstable
finite zero. Unfortunately, it can be shown that dependent control action is
necessary to move this zero to the left-half plane. I.N.A. analysis using the

dependent controllers 1

G,(s) = l-1 1]
1 -1
or [-;
---
s1 1s l! [-1.--. 1;• l
or
-----
1 +s
s -1-s
s
(2.5-8)

verifies system stability. However dependent control leads to very bad closed-
loop response and is thus not desirable. Therefore the best that can be

obtained is a simple control scheme designed to ensure a diagonally dominant

system for which single-loop theory can be applied to two loops that remain

1. Dependent control results from a singular compensator matrix.

stable only for moderate gains.

Table 2.5·2 summarizes the results for assuring diagonal dominance With

symmetric measurement placement. Since the conditions for stability and diag-

onal dominance are satisfied for z1 < 0.33, arbitrarily high gains can be applied

in each of the two principal loops Without instability. This is equivalent to the

results found in analyzing the multivariable root-loci. There it was shown that

for z1 < 0.33 high gains would not lead to instability; however, using the I.N.A.

method it is also shown that no attempt to eliminate interaction is necessary,

since large gains will lead to a system With little interaction. This can be seen by

considering the transfer functions of the process with a diagonal gain con-

troller, G0 (s) = diag (kt). The effect of applying a gain k, is to multiply qil and
qt 2 by k,- 1 . As k 1 and ~ are increased, the system approximates over an increas-

ingly wide band to a diagonal system. The transfer function between Yeti and y~

when the jth (j :t- i) loop is closed with gain kt and the ith loop is open leads to

(2.5-9)

As k 1 and k 2 become large, the interaction effect is small and

(2.5-10)

The design can then be carried out using the f'u (s) as though they were inverse

Nyquist diagrams for separate loops.

Finally it is instructive to consider the case where the measurements are

not placed symmetrically, i.e., z2 ;. 1. - z1 . The I.N.A. then consists of the four

diagrams representing qv(jc.>). Several cases were considered using the modified

Nyquist contour. Figure 2.5-6 shows the I.N.A. diagrams for the case (0.4,0.8).

which is diagonally dominant and stable. Thus arbitrarily high gains can be

applied, leading to a system With small interaction. Other cases that were
 -65-

o.oo < ~ < 0.)) cl1qoD&l cloainance and at&bill t.:r

uaured. ld.thout coapen.u.uon

a4d coapena&tor. t.o pt. dia.gor:tal

doainance and att..pt to control
\1118t&ble a1J:IIl• loopa

a4d perautat1on coapen.u.tor

cc • ( ~ ~I
to intercbaJ~se the a-..tr.rMenta
&Del. deaicn aa above

Table 2.5-2
I.N .A. Results for Symmetric Measurement Placement

considered are (0.4,0.7) and (0.3,0.8). Only the first of these requires compensa-
tion.
 -66 ~

Real

I mag

Figure 2.5-1
Diagonal Dominance Criteria
 -67-

0
0
LDii II
-3.00 -2.00 -1.00 9l.oo 1. 00 2.00 3.00 "· 00
-·· 00

(a)

Figure 2.5-2
Inverse Nyquist Plots for Outputs Zt =0.3, Z2=0.7
a. Logarithmic Plots
----lq 11 l --- ll+q 11 1 lq 12l
b. Element q 11 (s) c. Element q 12(s)
 -68-

c
c
N

OZ'£ o• ·z ag •t aa ·a 00 ciS aa ·a-

-
II: c

cQ
N
I
. -
'C)

c
.,..
c
I
't:l
~
;§
.
Q
Q
j;;l
»! CD 0
• u
N
I
~
c N
....c
c
. Q)

~
CD

cQ
.,..

--
J:l

-
II:

00 '9l 00
Q
Q

r.P
00 ··-
Q
Q

N
I
.
 -69-

g
0
.;

-.
0
ID

-----------------~

0
Q
LIIG II
-4.00 -3.00 -2.00 -1.00 '1:1.00 l. 00 2.00 3.00 •• 00

(a)
Figure 2. 5-3
Inverse Nyquist Plots for Outputs z 1=0.4, z2=0.6

a. Logarithmic Plots
----lq 11 1 ---11+& 11 1 --lq 1al
b. Element q 11 (s) c. Element q 1a(s)
 -70-

..
a
a

OO't oo·z di 0 oo ·z· 00 ··-

00 '9•
a ...
1:

..
a
a
I
--
(J

a
a
a)
I
't:1
Cl,)
a ='
a
. §
...
N
I
w c
0
(.)
C."'.)
I
a

..
It)
a N
a G)

~
w
a
a
c:i
N

-
1:

oo·sz oo·s 00
a
a
cF --
.0

a
a
Q
N
I

..
0
a
a
I
 - 71 -

Real

Imog

Figure 2.5-4
Modified Nyquist Contour
System with a Pole at the Origin
 -72-

0
g

•
0

"'...; I
I
/
..
0

...:;
//
I
0
I
-
10

0
Cll

-----------------~
0
0
LOG •
-t.OO -;,.oo -2.00 -1.00 '1l.oo 1. 00 2.00 3.00 •• 00

(a)

Figure 2.5-5
Inverse Nyquist Plots for Outputs z 1=0.4, Z2=0.6
a. Logarithmic Plots
----lq 11 1 ---11+q 11 1 - - lq 12l
b. Element q 11 (s) c. Element q 12(s)
 -73-

0
0
0
.
I •

g
0

oo·s 00 0 ·c;-
-
u
0
0
0
1"11
.
I

~
0 ~
..
0
0•
I
§
d
0
u
ID

.
0 t
0 10

I -
c.o
N
Cl)
r..

~
0
0
G
.

-
--
c
.1:1
oo ·zt oo·a oo ·a-
 -74-

0
0

-
ci

0
0
.;

0
0
.;

0
0
....

0
0
l'f

LOG •
--4.00 -3.00 -2. DO -1. DO 1. 00 2. 00 3.00 4.00

(a)
Figure 2.5-6
Inverse Nyquist Plots for Outputs z 1=0.4, z2 =0.B

a. logarithmic Plots
-----lq 11 1 ---lt+qul -lq12l
b. Logarithmic Plots
---- lq 221 --- ll+chel --lclell
c. Element q 11 (s) d. Element q 12(s)
e. Element q :H(s) f. Element Ct22(s)
 -75-

0
0
.;

/'

/
I
...
0

r.i

-.
0
0

/
/
I
_ .....
/
,A/
--......·
0
0 LOS II
-3.00 -2.00 -1.00 9:1.00 1. 00 2.00 3.00 (. 00

(b)

FigUre 2.5-6 Continued

 -76-

..,.
Cl
Cl

ot ·o ci 0 ot ·o- oa ·o- oz •t- 09 •t-

a
-•

--
Cl
a
...•
I
"0

a
0
.
"I
"0
CP
:::1
0
a
. .... :ad
-
N
I
c 0
(.)
co
I
1(.)
a N
a
IIi
i
oo·sc: .
-
c 0

0
0
In
.
I --
(,)

0
0
a...
I

0
0
....c ...•
In
I
 -77-

--....
0
(Q

...•
0
CD
0
.
"tt
~
d

-•
00'91 OO'Zl oo·a 00 ••
0
0

00 •cf
:;j
d
u0
00 ··- coI
t(.)
0
CD C\i
0
• Q)
....
~
09 'l oz 'l OQ'O Ot'O 00 dS ot ·o-
-• 0

.
--
0
G
Q)
t::t

-
co
•
I

I
0
....
N
.
I
 -78-

2.8 CHARA.Cl'ERI.S1'1C WCUS JIETHDD

The characteristic locus method (C.L.M.) is an extension of classical Bode-

Nyquist frequency-response theories and state-space techniques for the design

or general. multivariable feedback control schemes. The method combines the

essential features of both approaches by using the properties of linear vector

spaces defined over base fields of complex functions. As in the l.N.A. technique,
the C.LM. requires the use of a computer-aided graphic display and can be

incorporated into a computer-aided design package. Although the l.N.A. tech-

nique defines approximate conditions for stability using bands on inverse

Nyquist plots, the C.L.M. gives an exact indication of stability and a systematic

technique for choosing the best controller matrix in terms of system stability,

integrity, interaction, and accuracy, rather than simply stressing diagonal domi-

nance and single-loop design. The original method, developed by MacFarlane

and Belletrutti (Belletrutti and MacFarlane, 1971; MacFarlane and Belletrutti,

1973), has been refined and systematized through the late 70's (Edmunds and

Kouvaritakis, 1979; MacFarlane and Kouvaritakis, 1977) and has been experi-

mentally tested on an automatic flight control system (Kouvaritakis et al.. 1979)

and a two-bed reactor process (Foss et al., 1980).

The C.LM. uses the frequency dependent properties of the eigenvalues and

eigenvectors of the open-loop transfer function Q(s) = Gp(s)Gc(s) to assess the

closed-loop properties of the system. The design technique attempts to compen-

sate and modify the system's response by adjusting the eigenproperties of Q(s).

The basis for this use of the open-loop transfer function in the analysis of lhe

feedback system is the congruence of the closed- and open-loop eigenframes.

This relation can easily be seen from the dyadic expansions of Q(s) and R(s) for

unity feedback systems (MacFarlane and Belletrutti, 1973):

 -79-

(2.6-la)

(2.6-1 b)

where qi(s) are the eigenvalues of Q(s) and wi(s) and V;.T are the corresponding

eiR;envectors and dual eil'!;envectors. Obviously, the characteristic functions are

the same for both the closed- and open-loop systems and the eigenvalues are

directly related.

The analysis for multivariable systems leads to the following criteria for

achieving a compromise between the basic objectives of stability, integrity,

interaction, and accuracy (MacFarlane and Belletrutti, 1973). The basis of the

technique is the use of the characteristic loci, which are the paths in the com-

plex plane drawn by the eigenvalues of Q(s) as s traverses the standard Nyquist

contour.

Stability
Closed-loop stability can be assured by selecting the compensator

Ge(s) such that the net sum of the counterclockwise encirclements of

the critical point (-1.0,0.0) by the characteristic loci is equal to the

number of open-loop unstable poles Po

where m is the number of inputs and outputs of the system and is

thus the number of loci needed for complete description of the

system's eigenproperties. It should be noted that the direction along

the characteristic loci is taken as that which corresponds to a clock-

wise traversal of the Nyquist contour. For the third-order, lumped

model of the heat conduction process, the open-loop characteristic

 .. 80-

equation is s(s + rr 2 )(s + 4rr2 ). Since the poles are al 0, -rr 2 , and -4rr 2 ,

p 0 ::: 0, and the encirclement criterion reduces to

(2.6-3)

Integrity

A multivariable feedback system is of high integrity if it remains

stable under all types of failure conditions. The major probable

difficulties include output transducer, error-monitoring channel. and

actuator failures. For practical systems, the control design must

obviously take into consideration such component breakdowns.

Integrity against transducer and error-monitoring channel failures in

all possible combinations of the loops can be assured if the charac-

teristic loci of all the principal submatrices of Q(s) satisfy the encir-

clement criterion. For integrity against actuator failures, similar con-

siderations apply to the matrix Gc(s)Gp(s).

Interaction

Reduction of interaction in a multivariable system is usually desired

to improve closed-loop dynamic response and is often necessary to

assure stability. At low frequencies, interaction can be suppressed by

imposing high gains, i.e., lqi(jr.>)l > > 1.0. Then the dyadic expansion of

R(s),

(2.6-4)

shows that the closed-loop system becomes essentially non-

interacting. However, this restriction cannot always be achieved

because stability conditions usually require that the characteristic

 - Bl -

gains have small moduli at high frequency. Since the eigj';'lnvalues of

G(s) satisfy lqi(jc.>)l < < 1.0 at high frequency,

R(jc.>) -* ~ ~(jc.>)w,(jc.>)vl(jc.>) = Q(jc.>) ,

1::1

and therefore the cross-couplings in R(jc.>) are carried over from Q{jc.>).

Thus to suppress interaction at high frequency, Gc(s) should be

selected so that the eigenvectors of Q(s) are nearly as possible aligned

with the standard base set. If the measure of alignment is taken as

the angles between the eigenvectors w,(jc.>) and the base vectors ~ for
i= 1,2 (since m =2 for the heat conduction system), the objective is
to reduce the misalignment angles. MacFarlane and Belletrutti ( 1973)
show that, although these criteria lead to a reduction in interaction,

they do not necessarily result in a Q(s) which is nearly diagonal or

diagonally dominant, and thus a feedback system can be made nearly

non-interacting without imposing diagonal dominance.

Accuracy

A system has high accuracy if the actual system output closely follows

the desired output. In e;eneral, system accuracy will be high provided

that the characteristic loci have large moduli at low frequency.

The objective of the characteristic locus method is then to select a con-

troller Gc(s) so that the characteristic loci of Q(s) satisfy the stability criterion

and have high gains at low frequencies and low misalignment angles at high fre-

quencies. Additionally, Gc(s) should be selected so that its elements are rational

functions of s, so that IGc(s)l is nonsingular and has no right half plane zeros,

and so that aU the poles or Gc(s) are in the open left half complex plane.
 -82-

MacFarlane and Kouvaritaki.s ( 1977) present a systematic approach for

using the C.L.M.. In particular, a means of manipulating the characteristic loci

and characteristic directions is developed so that the phase of the loci can be

adjusted to achieve acceptable stability and integrity and so that the directions

can be aligned and the gains balanced to reduce interaction. As outlined in

several publications, a procedure for optimal alignment of a given complex

plane with a real frame according to a misalignment measure can be incor-

porated into a computer algorithm. 1 The design procedure can then be split into

two distinct parts:

i. Htgh-Frequency Controller Performance

At high frequencies, it is desirable to reduce the misalignment angles

between the compensated system's characteristic direction set and

the standard basis vectors. This can be accomplished by designing a

real compensator KH that approximates the complex frame of Gp1 (j(.))

at some high frequency i:Jb.. A program ALIH which was written for this

purpose uses the routine ALlGN to perform the actual alignment. A

listing of the program is presented in Appendix 1. The signs of the

columns of KH are arbitrary and are selected so that the eigenvalues

of ~ are positive. real. The compensated system then has infinite

zeros with asymptotes on the negative real axis.

ii. Low-Frequency Controller Performance

At low frequencies, the encirclement criterion should be satisfied and

the moduli of the characteristic loci should be large. This can be

accomplished by manipulating the loci with an appropriate approxi-

1. This algorithm (ALIGN), which is a basis for high frequency alignment and low frequency manipu-
lations, was written specifically for this project and is included in the programs AI.JH and ALIL
d.isoussed below and listed in Appendix 1. It should not to be confused with the program AI.JGN
referred to by other authors, although their structures should be similar.
 -83-

ma.tely commutative controller Ki.. where JG. = Af'k(s)B with A and B

being the real frame matrices that approximate the complex frame of

the eigenvector matrix W(jc.J) and dual eigenvector matrix V(jCJ). Again

the rout.i.ne ALIGN is incorporated into a program AUL for this pur-

pose. This program is also listed in Appendix l. The elements of the

diagonal matrix rk(s) are then chosen on the assumption that q.,; =
g..;'!<i where q,: and g, are the eigenvalues of the compensated and

uncompensated system and k. are the elements of rk(s). The compen-

sator Ki. is thus used to insure stability and integrity and to adjust the
gains at low frequency to reduce interaction.

The etiect of I<H and KL must be combined in such a manner so that each

operates in the appropriate frequency range without significantly altering the

efiect of the other in its appropriate range. This is achieved by using the low-

frequency controller as matrix-integral control:

Gc(s) = ~KL + I<H (2.6-6)

s

where a is a constant chosen to achieve a suitable transition from the low-

frequency to high-frequency behavior of Ge(s).

Although the C.L.M. is seemingly complicated, it is actually a simple design

technique once the appropriate computer facilities and programs are available.

The major programs are ALIH and AlJL. while other programs were used to cal-

culate and graphically display the characteristic loci and misalignment angles.

All programs were tested on the automatic flight control system (Kouvaritakis et

al., 1979) and gave results equivalent to those published. Then for the heat con-

duotion process, the design was performed using the third-order, lumped model

with two controls and outputs and then with an extra measurement, which was
squared down in an inner loop.
 -84-

2.6.1 Basic System Analyais

For the uncompensated system, characteristic locus plots were drawn for

the cases that were considered in the I.N.A. design (Table 2.5-1). Figures 2.6-1 -

2.6-3 show the loci With the measurement locations (0.3,0.7), (0.4,0.6) and

(0.4,0.8). 2 For the symmetric cases with z1 < 0.33, the encirclement criterion is

satisfied for all gains. As was also shown using root-loci and I.N.A. analyses,

these cases Will remain stable with increasing gain. Also as expected, high gains

lead to instability for 0.33 < z 1 < 0.50. In particular, a maximum gain of 32.0

would be allowed before the critical point is encircled tor z 1 ::: 0.4. Thus, for the

symmetric cases, compensation is required for stability of the system at high

gains when 0.33 < z1 < 0.50. Additionally, regardless of the measurement loca-

tion, one eigenvalue has large magnitude at low frequencies while the second has

very small magnitude, and the misalignment angles are both 45° at high fre-

quencies. Thus some compensation is desired for all the cases to improve

integrity, interaction, and accuracy of the system. Although the unsymmctric

case (0.4,0.8) is also stable for all gains, compensation is desired to improve its

closed-loop response. Since the cases (0.3,0.7), (0.4,0.6), and (0.4,0.8) are

representative of the problem, they will be analyzed in detail using the C.LM ..

2.6.1.1 Ca.se 1: Measurements z1 = 0.3, Z2 = 0.7

As discussed above, this case satisfies the stability criterion for all gains.
The objective of the control action is thus to improve dynamic system perfor-

mance by aligning the characteristic directions at high frequency, balancing the

gains of the characteristic loci at low frequency, and injecting gain into the

overall system. while maintaining stability. An analysis of the integrity against

2. Note that! the ver.tlical axis in the figures of lq"l versus frequency I'U'e defined in terms of
dB =20 los q1. (j r.>) .
 -85-

all types of failures leads to the conclusion that the system is of high integrity,

since the characteristic loci of the principal subrnatrices of G(s) also satisfy the

encirclement criterion.

Preliminary analysis shows that the compensator

(2.6-7)

balances the gains perfectly and significantly reduces misalignment, but further

analysis shows that such a compensator leads to system instability. Therefore

rather than trying to guess the appropriate control structure, it is best to per-

form the systematic alignment procedure discussed previously.

The first step involves designing KH to reduce the misalignment angles. For

the third-order model of the heat conduction with symmetric measurements,

the matrix of eigenvectors is approximately8

1.0 + Oi -1.0 +Oil (2.6-8)

W= [ l.O+Oi l.O+Oi

throughout the frequency range 10-4 to 10~ F1gures 2.6-1 and 2.6-2 show that as

expected the angles between the standard base vectors and the characteristic

direction set are about 45°. . The alignment is attempted using ALIH at several

frequencies in the range !'.>}1 = 1 .... 100, but unfortunately the analysis of many

possible compensators :Kh leads to the conclusion that no real matrix can ade·

quately reduce misali.inment angles at high frequencies for symmetric measure-

ments without leading to system instability. This problem results from the

eigenvector matrix W being nearly real and having off-diagonal terms identical in

magnitude to the diagonal terms. Although it seems plausible to select a com-

plex compensator KH = G- 1 (jr.>)l), the resulting controller is not physically realiz-

3. The imaginary parts are of 0(10- 10) at cw =10-4 and O(lQ-4) at cw =104).
 -86-

able. Thus KH should merely be selected as the identity matrix. If in~tead KH is

selected as the zero matrix, Gc = E!..
s
:KL and as s approaches infinity, Gc
approaches zero. This could lead to problems.

At low frequencies, AlJL is designed to calculate real matrices A and B which

approximate the eigenvector and dual eigenvector matrices; however. this leads

to numerical difficulties with symmetric measurement placement. The problem

is actually trivial since W is approximately real at all frequencies (especially at

low frequencies). Thus with the selection of A= Wand B = V. Ki:. = WAkV. The

gains were then balanced at the frequencies ">>.. = 1.0, 0.1, 0.01, and 0.001 by

selecting

[1OJ
A= 0 1 ,
lq 1(jCJL)I
where 1 = l~(jCJL)I (2.6-9)

Furthermore since Gc(s) = E!..Ki. + I. the values of a can be adjusted to give the
s
best response. Mter consideration of many combinations of a and (.)L, the best
balance at low and moderate gains was obtained with

10[-5.4
Gc(s) = ~
6.4 -5.4J
6.4 + [10 OJ1 (2.6-10)

Figure 2.6-4 shows that this compensator leads to a stable system with nearly

identical gains at c.>= 1.0 and high gains throughout the range c.>< 1.0. However,

the misalignment angles have not been reduced. The final step in the design is

to increase the moduli by injecting gain into the compensated system.

2.6.1.2 Case II: Measurements z1 =0.4, Z2 =0.6

Since this case is stable only for gains less than 32.0, the major concern of
the compensator should be to insure stability for higher gains. Attempts are
 -87-

made to design such a high frequency compensator. However even by using A1JH

and selecting the compensators so that the eigenvalues of CBKH are positive, the
overall system remains unstable. This is explainable since the condition that

the eigenvalues of CBKH are positive is necessary for stability but is not
sufficient. In addition to the two infinite zeros, there is a finite zero at approxi-

mately +63.0, according to the root-loci analysis. In designing K.H. there are only

two degrees of of freedom available for selecting the appropriate signs, so only

the two infinite zeros can be placed. Further analysis shows that dependent
control action is necessary to move the finite zero into the left half plane.4 The

problem essentially reduces to a single-input, double-output system. However

since dependent control leads to very bad closed-loop response/' the best that

can be accomplished is to reduce interaction and mcrease accuracy Wlthout

eliminating the instability for high gains.

Using AIJH and Al.JL, the best design for balancing gains and reducing

misalignment angles is

1 0 [ 11.8 -10.8] + [ 0 -1]

Gc(s) = ~ -10.8 11.8 -1. 0 (2.6-11)

The characteristic loci for the system with this compensator is shown in Figure
2.6-5. Unfortunately, this system is unstable for all gains and is therefore not

desirable. Thus a better control strategy is necessary for the heat conduction

system with measurement locations (0.4,0.6).

2.6.1.3 Case III: Measurements z1 ::: 0.4, Z2 =0.6

Though this case is stable for all gains without compensation, Figure 2.6-3

shows that a compensator may be quite useful to balance the magnitudes of the
4. The same conclusion was reached using the I.N.A. design procedure.
5. This is shown in the simulations in Section 2.7.
 -68-

loci at low frequency and to reduce misalignment angles at high frequency.

Whereas the misalignment angles were always 45° and could not be reduced in

the symmetric cases without causing instability, high frequency compensation is

useful for this unsymmetric case. Considerable analysis using AlJH and AlJL at

various frequencies along with adjusting a leads to an excellent C.L.M. design.

With the compensator

0.8 [ 4.348 -4.348] [1.0 -Q.371]

Gc(s) = -s- -4.348 4.348 + 0 0.929 · (2.6-12)

the characteristic loci (Figure 2.6-6) have nearly identical moduli at low tre-

quency, and the misalignment angles are quite small at high frequency. Thus

the C.L.M. seemingly leads to a stable system with high integrity, low interaction

and high accuracy. Furthermore, the method concludes that this design has the

best compromise between these qualities, since the compensators were designed

using approximately commutative theories. The final step in the design is to

inject gain into the compensated system.

Thus the C.L.M. provides a SJ5tematic approach for designing a system

which best satisfies the criteria of stability, large moduli of the characteristic

loci at low frequencies and small misalignment angles at high frequency. Simu-

lations are necessary to show the actual extent of the improvement obtained by

using this design technique, but it is apparent that the designs for the sym-

metric cases do not adequately satisfy the C.LM. criteria. Only for the (0.4,0.8)

case is an excellent design obtained. Thus it may be advantageous to use extra

temperature measurements within an inner loop to improve the system, for

which the C.L.M. will be used to design the outer-loop controller.

 -89 . .

2.6.2 Inner-I.Dop Analpia

The use of easily available measurements in excess of the commanded out-

puts6 is considered, so as to make more effi.cient use of the gain. The procedure

involves squaring down the extra measurements within an inner loop, in order

to form a new set of outputs equal in number to the number of inputs. Instead

of using the extra degrees or freedom to reduce interaction as was done in Sec-

tion 2.4, the objective is to use the inner loop to suitably place the poles of the

outer loop so that the C.L.M. (or I.N.A. method) can be applied to a 'better' sys-

tern as seen !rom a control point of view. The inner-loop design is based on the

placement of the finite zeros and the manipulation of the root-loci asymptotes

for the inner loop so that by finally setting the inner-Loop gain at a suitable
value, the poles of the outer loop are pulled into better locations in the complex

frequency plane.

Consider the process shown in Figure 2.6-7. The outer loop consists of two

measurements and two controls. However, three measurements are actually

taken and squared down in an inner loop usill€ the compensator F. The system

is described by the following relationships, where Land C were defined in Section

2.4.

y = J£x = C:x: .YF=h="FCI

I= (sl-A)- 1Bu
a = Gc[yd-y] .
u =Ki:(a-YF)
The method involves designing F. from which Y can be calculated if C is non-

singular. Using the above description of the process, the inner-loop closed-loop

response is

(2.6-14)

6. The commanded outputs are those directly needed for comparison with the set points.
 -90-

The characteristic equation for the inner loop is

lsi-AI·II+GtKil = o . (2.6-15)

from which the root loci are obtained by letting Ki = kl Then the overall closed-

loop response is given by

with Gp given by Equation (2.2-27), and the characteristic equation for the

overall system is

(2.6-17)

Using Gc(s) = k'l, the poles of the outer system are given by k' = 0 and are thus

described by the solutions to

lsJ-AI·II+GtKil = o . (2.6-18)

Since Equations (2.6-15) and (2.6-18) are equivalent, the poles of the outer sys-

tem are described by the root-loci or the inner system. Thus the inner-loop

design can be used to place the outer-loop poles by selecting k.

Consider the case with F = C. 7 Then since G:t = Gp. the root-loci for the inner
loop are identical to those obtained in Section 2.2 for the outer loop. Thus if the

root-loci of the original system without an inner loop show good characteristics-

-zeros well into the left half plane and little oscillation--then extra measure-

ments may not be needed. The outer-loop poles can be shifted simply by imbed-

ding an inner loop with sufficient gain. This is similar to the ideas of modal con-

trol.

The desired characteristics of the inner root-loci are that the infinite zeros

are along the negative real axis and that the finite zeros are well into the left

half plane. Obviously. many possible Ji' matrices can lead to such root-loci. In
7. One possibility would be to take no additional measurements.
 - 91-

the following analysis, we consider only four such matrices.

Remembering that the root-loci of the original system with z1 < 0.33 and z2

= 1.0 - z 1 satisfy the desired criteria, the inner-loop compensator can be taken
as F =C(z1,z2). where C(Zt.Z2 ) is the C matrix obtained with the points z1 and z2 . 6
The cases that were considered are FA, FB. and Fe shown in Table 2.6-1.

,.
A
( 1.000 o.tt)l
1.000 -1.141+
-o ...,.,)
0.4)7
• C( .J•• 8) It • B.O

, . ('.000
B
0.8)1.
-··"11)
1.000 -0.8)1. -0.4)7
• c(.J•• ?} k. 6.0

,c . ( I.OO<l 1.144
1.000 -1.144
•·•11) •
0.4)7
C( .2 ••8) Jt • 10.0

(.....
'n • z.ooo
1.000 .......) k. 4.0
-1.000 -o.414

Table 2.6-1
IQner-Loop Compensators

For F rt: C, the criteria necessary to have infinite zeros along the negative

real axis (i.e., FB has all positive, real eigenvalues) and the finite zero well into

the left half plane are

8. Note that the Z1 and Z2 have no relation to the actual mealJW'ement locations for this inner-loop
problem. This structure of F :is only used for convenience since we know that the C matrices
=
with z1 < 0.3_3 and Z:z l.Q..- z1 have root-loci with the desired features. The actual measure-
ments define C, from which F is determined.
 -92-

a) fu + ~fuz + ~fts > 0

b) (ft2f21-fufr 22 ) + ~(f2sf12-f22fts) > 0

c) f2tfu! - fuf22 >> 0 .

Although the elements of F should be selected to satisfy these conditions, there

are still an infinite number of possibilities available. However, consider

- [ 1
B= ~ -v2"
1 l a.= fu +~f13
p = f21 +"V"2'f23
(2.6-19)

and specify that Q =FB =FB should have all real, positive eigenvalues. Then

with the further restriction that the rate of divergence to the infinite zeros is

the same for all corresponding root-loci. F must be of the form F =7s-1. After

trying various values of -y, f 11 , and f 21 , the best inner-loop design using this tech-

nique was found to be Fo shown in Table 2.6-1.

The second stage of the design procedure, the outer-loop design, consists of

the application of the C.L.M. to the system with the inner loop in place. Again

the design is considered for commanded outputs (0.3,0.7), (0.4,0.6), and

(0.4,0.8). Since the inner loop has been designed and since only z 1 and z2 are

needed for the outer loop, z3 can be selected arbitrarily as long as C remains

nonsingular. This restriction is satisfied if Za is not equal to z 1 or z2. Then

although the best C.L.M. design for the system without an inner loop was unable

to reduce misalignment angles, it may be possible to improve on that design.

2.6.2.1 Case I: Mea.suremants z1 = 0.3, z2 =0.7

Although the outer-loop design was performed using all the inner-loop com-

pensators shown in Table 2.6-1, only the analysis With FA will be discussed since

all the cases lead to similar loci and since FA leads to the best overall design. A
 -93-

preliminary examination of the characteristic loci for this case without outer-

loop compensation (Figure 2.6-8) establishes the need for such compensation

due to the disparity of the characteristic gains and the high misalignment

between the characteristic directions of Q(j~) and the basis vectors.

The C.I...M. is then performed using slightly modified versions of the pro-

grams previously discussed. Although AUH can be used to calculate .KH. un!or-
tunately it cannot be used to solve the inherent sign ambiguity. PreViously this

was possible since the infinite zeros were along the negative real axis if CBKH had
positive, real eigenvalues. However, the analysis with the inner loop in place

shows a much more complicated result. If r. C, and B were square matrices, the

simpler result could be derived.

With the inner-loop structure, it is possible to change the misalignment

angles by designing an appropriate .KH. However whenever one of the misalign-

ment angles is reduced at high frequency, the other is increased by about the

same amount. Thus KH =I still provides the best compromise. Low-frequency

compensation is then designed using AUL at several frequencies, and a is

adjusted to give the best balance between the low and high effects. The final

overall design is

F = [1.0 0.831 -Q.437]

JG = diag(8.0) 1.0 -1.144 0.437
(2.6-20)
10.0 [ 1.267 -Q.267] [1 OJ
Gc(s) == -;-- -Q.457 1.457 + 0 1 ·

The characteristic loci With this design are shown in Figure 2.6-9. They obviously

indicate that this design is better than that without an inner loop. In particular,

the loci are much better balanced, and although the misalignment angles were

immobile Without an inner loop, they are greatly reduced for moderate rrequen-
 -94-

cies with this design. Unfortunately, they are still 45° at very high frequency

(c.>> 100 ).

2.6.2.2 Case II: Measurements z1 =0.4, z2 = 0.6

Previously this case was found to be unstable for high gains due to an

unstable finite zero, and it was shown that only dependent control would lead to

stability since such control action was necessary to move the finite zero. Unfor-

tunately. simple algebraic calculations show that an inner loop has no effect on

the finite zeros of the system, and therefore the stability arguments are

unchanged. Nevertheless, a compensator can be designed to improve the sys-

tern by reducing interaction and increasing accuracy, if the values of the gain

are restricted so as to assure stability.

Figure 2.6-10 shows characteristic loci for the best overall design that was

obtained using the C.L.M.. The design consists of

Ki: = diag(8.0) F =[1.0 0.831 -Q.437]

1.0 -1.144 0.437
(2.6-21)
1.0 [1 OJ [ 0 1]
Gc~s) = -s- 0 1 + -1 0 ·

This system has a good balance of gains at low frequencies and low misalign-

ment angles at high frequencies but will be unstable for outer-loop gains greater

than 3.1. Since this is an overall gain of 24.8, 9 it is not much lower than the sta-

bility limit for the uncompensated system.

9. The overall gain is the product of the outer-loop gain and the inner-loop gain.
 -95-

2.6.2.3 Case III: Mea.s'I.I:T'emrmts z1 = 0.4, z2 = 0.8

Since the C.L.M. design for this case without an inner loop led to a compen-

sator that perfectly balanced the gain at low frequency and significantly

reduced the misalignment angles at high frequency, there seems to be little or

no need to add an inner loop. Nevertheless, such an analysis was performed

leading to the following design.

Ki = diag(8.0) :r =[1.0 0.831 -Q.437]

1.0 -1.144 0.437
(2.6-22)
- 10.0 [1.080 -0.212] [1 0]
Gc (s ) - s 0.091 0.747 + 0 1 .

The corresponding root-loci are shown in Figure 2.6-11. This design does not

appear to be as good as that without an inner loop. Further refinement could

lead to a slightly better system.

2.6.3 Modified Inner Loop Analysis

Finally, another outer-loop design suggested by Kouvaritaki.s et al. ( 1979)

was considered (Figure 2.6-12). Since the purpose of the inner loop is to

improve system dynamics oy making use of all available measurements, while

the overall objective is to exercise control over two of the measured variables,

some minor modifications are made so that the inner loop operates only on the

extra measurement. The new configuration makes more efficient use of gain

since il closes lhe loops around z1 and z2 only once.

The design consists of

41 ;: (0 0 1) (2.6-23)

while F2 is 2 x 1 and F1 is 2 x 2. However using this design, the inner-loop root-

loci for the previous system (Figure 2.6-8) do not describe the behayior of the

poles for the outer loop of the new system. Only with Gc = I are the resulting

root-loci for varying k identical to those obtained before rearrangement. Apply-

ing the C.L.M. to the new inner-loop system results in similar characteristic loci

to those obtained previously, but the new configuration makes more efficient

use of the gain.

Finally it can be concluded that, although the inner loop is often beneficial,

it does not always improve the overall design. If the criteria for a good C.L.M.

design can be satisfied without an inner loop, then further analysis may not be

necessary. However, if the misalignment angles cannot be reduced or the gains

balanced without an inner loop, the inner loop may be useful.

 -97-

(.,)

-0. 3:: -fl. 2C -C. I C ':-L:l. 00 0 10 0.20 0.30 c..w

'""
0I

...,
0

.,
0

c:iI

0 IM
1,')

:9

(a)

Figure 2. 6-l
Characteristic Loci Diagrams for Outputs z 1 =0.3, zz=0.7

a. C.L.M. Plots b. Magnitude vs. Log Frequency

c. Misalignment Angles vs. Log Frequency
 -98-

c
0
c::i /Q/,!11!
co

0
0

-4. 00 -3.00

0
0

,...ci
'
(b)
0
C)

c::i TMETq
"'
0
0

...
d

0
0

,c.i
0
0
c::i
N

0
0
c::i

0
0

-4. 00 -3.CO -2. co -1. 00 9:l. co I. 00 2. co 3. co 4.00

(c)

Figure 2.6-1 Continued

 -99-

.....
-
0

c 1 ~

""c:;>

(a)

Figure 2.6-2
Characteristic Loci Diagrams for Outputs z1=0.4, z2 =0.6

a. C.L.M. Plots b. Magnitude vs. Log Frequency

c. Misalignment Angles vs. Log Frequency
 ~ 100 ~

0
0
c:i 101, OS
'""

0
0
c:i
"'
0
0

-3. 00
-·· 00

(b)

0
0
c:i T~Erq
Ill

0
0

....0

0
Q

..,c:i
0
0
c:i
"'
0
0
c:i

0
0

-3.00 -2.00 -1. oa cn.oa l. 00 2.00 3. co ... co

-·· 00
(c)

Figure 2.6-2 Continued

 - 101 -

-o. OS 0 00 o.os o. 10 o. 15 0.20 0.25 o. 30

-
Ill

0I

....,
""0
I

If)

'"
0 I

....,
... Ill
0
'

(a)

Figure 2.6-3
Characteristic Loci Diagrams for Outputs z 1=0.4, z2=0.8

a. C.L.M. Plots b. Magnitude vs. Log Frequency

c. Misalignment Angles vs. Log Frequency
 - 102-

0
0

-c:i
0 10/,01

LDI II
~ •• 00 -3.00 -2.00 -1. 00 2.00 3.00 4.00

0
0
c:i
Vl
I

0
0
c:i
0
7
0
0
c:i
In
7

{b)
0

gi""'
0
0

...
c:i

c
0

..,
c:i

0
0
c:i
N

0
c

-
c:i

0
0 LOf M

-·· 00
·3. 00 -2.00 -1.00 '11. 00 1. 00 2.00 3. 00 4.00

(c)

Figure 2.6-3 Continued

 - 103-

lJ'l
N

-0 . .I,C! -C. 30 -c. 2c -c. 1c c:i.. • co 0. lC c. 20 C.30

lJ'l
':r:t

0
C1'

':'

If)
a'

'?

c.J
C..> !'I
,;

(a)

Figure 2. 6-4
Characteristic Loci Diagrams for Outputs z 1=0.3, z2=0.1
Compensator Given by Equation (2.6-10)

a. C.L.M. Plots b. Magnitude vs. Log Frequency

c. Misalignment Angles vs. Log Frequency
 - 104-

0
0
0
... 1:11,08
"'
0
0

0
0
0
G)
I

0
0
0
II)
7

(b)
0
0
0
,,., HtEHI

0
0
c:)
....

0
0
0
...,

0
0
0
N

0
0
0

0
0

-4. 00 -3. 00 ·2. co -1. co 9l. co I. 00 2.00 3. 00 4.00

(c)

Figure 2.6-4 Continued

 - 105-

-0.~0 -0. 15 -0. 10 o. 10 0. 15 0.2()

....0
ciI

...
'::>
I~

c?

(a)
Figure 2.6-5
Characteristic Loci Diagrams for Outputs z1=0.4, z2 =0.6
Compensator Given by Equation (2.6-11)
a. C.L.M. Plots b. Magnitude vs. Log Frequency
c. Misalignment Angles vs. Log Frequency
 - 106-

0
0
c:i
1~1. OS
"'
~,

0
0

...
c:i
"'
0
0
c:i
<0

0
0
c:i
a:;

0
0
c:i
II?

(b)

0
0
c:i T'fET~

"'
0
0
c:i
~

0
0

,.,
c:i

0
0
c:i
N

0
0
0

0
0 to~ ..
-~.co -~.co -2. co -1. co en. co I. 00 2.00 3. co 4. co
(c)

Figure 2.6-5 Continued

 - 107-

0
-0. 10 -0. OS 9J 00 0. OS 0. 10 0. 15 0.20 0. 25

...
0

c:iI

..,.0
c:iI

0
tO
c:iI

0
()) I"
c:iI

(a)

Figure 2.6-6
Characteristic Loci Diagrams for Outputs z 1=0.4, z2 =0.8
Compensator Given by Equation (2.6-12)

a. C.L.M. Plots b. Magnitude vs. Log Frequency

c. Misalignment Angles vs. Log Frequency
 - 108-

c
c
0c
- /0/,01

c
c
0
In

UG II

-<&.00 -3.00 -2.00 -!. 00 2.00 3. 00 •• 00

c
c:>
0
11'1
I

g
0
c
i
0
0

-0
11'1
I

(b)
0
0
o
...., ThUll

0
c

-
0

0
0 lGG II

-<&.00 -3.00 -2.00 -1.00 93. 00 1.00 3. 00 •• oo

(c)
Figure 2.6-6 Continued
 - 109-

Yd + a +..:..,. u YM y
'<>'_ * Controller
Gc ~>'- Kx r--- Process
Gp L

YF F

Figure 2.6-7
Inner-Loop Control Design Strategy
 - 110-

0
I'll
d

0
-0.20 9l 00 0.·2D. o•. so. 0.-80 1. 00 1. 20
1111:

0
ID
dI

gd '"
I

(a)
Figure 2. 6-8
Characteristic Loci Diagrams for Outputs z 1=0.3, z2 =0.7
Inner-Loop Compensator FA
No Outer-Loop Compensation

a. C.L.M. Plots b. Magnitude vs. Log Frequency

c. Misalignment Angles vs. Log Frequency
 - 111 -

Cl
Cl
d /1/,01
"'
Cl
Cl

Cl
Cl
d
Cl
I

(b)

Cl
Cl
d
on
THUll

Cl

.
0
d

Cll
0

....

0
0
d
"'
0
0

-
d

0
0 LOt II
-3.00 -2.00 -1.00 91.00 1. 00 2.00 3.00 4.00
-·· 00
(c)

FigUre 2.6-8 Continued

 - 112-

0
N

-o.eo -o.ss -o.!O 0.45 0.70 0.95 1. 20

-.
I

(a)

Figure 2. 6-9
Characteristic Loci Diagrams for Outputs z 1=0.3, z2 =0.7
Inner-Loop Compensator FA
Outer-Loop Compensator Given by Equation (2.6-20)

a. C.L.M. Plots b. Magnitude vs. Log Frequency

c. Misalignment Angles vs. Log Frequency
 - 113-

0
0
c:i
-
N 1111.01

0
0

.,c:i

Cl
0
c:i
....

-1.00 <1J 00
0

..
0
c:i
I

Cl
Cl
c:i

' (b)

0
0
c:i
II)
THnll

LOG II
-&.00 -3.00 -2.00 !.00 •• 00

(c)

Figure 2.6-9 Continued

 - 114-

....0

-o. eo -0.55 0. 4~ 0. 70 0.95 l. 20

(a)

Figure 2.6-10
Characteristic Loci Diagrams for Outputs z 1 =0.4, z2 =0.6
Inner-Loop Compensator FA
Outer-Loop Compensator Given by Equation (2.6-21)

a. C.L.M. Plots b. Magnitude vs. Log Frequency

c. Misalignment Angles vs. Log Frequency
 - 115-

0
0
ci
o 101. DB

0
0
ci
0
'7
0
0
ci
J'l
-;-

(b)
0
0
ci l'IEl~

"'

A
....

0
0
ci
,..,

0
0
ci
I"J

0
0
ci

0
0

-4. co -j. co -2. 00 ·I. 00 'il. co I. co 2. 00 3. 00 ... co

(c)

Figure 2.6-10 Continued

 - 116-

(a)

Figure 2.6-11
Characteristic Loci Diagrams for Outputs z1 =0.4, z2 =0.8
Inner-Loop Compensator FA
Outer-Loop Compensator Given by Equation (2.6~22)

a. C.L.M. Plots b. Magnitude vs. Log Frequency

c. Mi:salignment Angles v:s. Log Frequency
- 117-

0
0
ci
CJ 101.01

0
0
d
.n
 - 118-

yd +..10\ • y- y
+,o,. Process
.>:
K(S) Fi '< ,>'_ kl ~ Gp ~pt.. L,
-
yf
F2
.__ L2 1-

Figure 2.6-12
Alternate Inner-Loop Control Strategy
 - 119-

2.7 CONTROL SYSTEM PERFORIIANCE

The control structure analysis for the heat conduction system was per-

formed on the third-order, lumped model using various frequency-response

techniques. These include non-interacting control methods, root-loci analysis,

inverse Nyquist array, and the characteristic locus method. Additionally, the

time-domain techniques of optimal and modal control were applied to both the

lumped and distributed models. Since all of the schemes provide a design for

the control structure, a comparison among the system responses is needed to

determine the extent of improvement in the closed-loop behavior of the overall

process. Additionally since most of the designs were performed using the third-

order lumped model, simulations of the actual system are necessary to assess

the effectiveness or the designed controllers. Computer simulations or lhe sys-

tem responses to various set point changes were performed for the different

control system structures. Simulations of the optimal control strategies were

nol performed, since such simulations could not easily be compared to those

using frequency-response designs due to the necessity of defining the conditions

of optimality. In practice, these conditions are usually selected to minimize

energy costs or to increase profit margins, rather than simply to reduce

response times or interaction. Responses of the heat conduction system with an

optimal controller to set point changes and system disturbances are shown in

literature (Betts and Citron. 1972; Sakawa. 1964).

Simulations of the heat conduction process can easily be performed using

the modal or laplace models derived in Section 2.2. Since both of these were

shown to be equivalent, either model can be used; however, since the frequency-

response designs were performed on the lumped, state-space representation

obtained from the modal-lumping procedure, the simulations used the following

results:
 - 120-

(2.7-1)

1.0 n=O
'i"n(z) = [ ~cosnrrz n = 1, 2, ...
(2.7-2)

with an(t) given by

(2.7-3)

1
an(O) = fo 9'n(z)yo(z)dz ,

n=O
n = 1. 2, ... (2.7-4)

Then if y0 (z) =0,

(2.7-5)

The closed-loop response y(z,t) can then be obtained by applying a discrete time

analysis to the control structure and feedback information. Although the actual

distributed system is described by the infinite-order model (Equation 2.7-1),

simulations show that a tenth-order analysis accurately describes the actual

process. Thus system responses for the various control strategies were calcu-
lated using both third- and tenth-order lumped process models.

The first control structure to be considered is proportional feedback con-

trol. Although such control may seem excessively elementary, it has the advan-

tage of simple implementation and is the method of control recommended by

several of the design techniques. For stable, diagonally dominant systems, the

conclusion of an I.N.A analysis is that no attempt should be made to reduce

interactions by using more complicated compensation. Instead, large gains

should be used in each feedback loop, leading to a stable system with little
 - 121-

interaction. The root-locus analysis, C.L.M., and I.N.A. methods show that the
symmetric cases with z1 < 0.33 and the unsymmetric cases (0.3,0.8) and

(0.4,0.8) are diagonally dominant and stable. However since these conclusions

were based on the third-order lumped model, proportional control should be

attempted on the actual system.

In general, the simulations show that the exact process behaves qualita-

tively similarly to the third-order model and that high gain significantly reduces
interaction for certain cases. However, proportional control leads to consider-

able offset. Even if only one of the two set points is changed, there is offset in

both outputs. This is obviously a result of the steady state interactions in the

system. Although steady state decoupling would seem beneficial. previous

analysis showed that such compensation leads to a dependent control system.

Thus although the root-loci and l.N.A. analyses showed that arbitrarily high
gains would reduce interaction without leading to instability in certain cases,

very hlgh gains are often needed to sumciently reduce the otrset and interac-

tion. Such high gains may be impractical or may lead to large overshoot, and
even when steady state interactions are reduced by high gains, considerable
dynamic interaction may still be present.

Thus although simple introduction of high gain into both control loops may
be adequate, the control structure can often be improved by a slightly more

complicated compensator. From classical theory, it is obvious that some form

of integral action should be introduced to eliminate the offset. The C.L.M. deter-

mined the best compensator of the proportional-integral (PI) form to reduce

interaction, to increase system integrity and accuracy, and to insure stability.

However, simple analysis shows that, due to the integral action, the closed-loop
responses become oscillatory. Simulations were performed using the C.L.M.

destgns With and without the inner loop. They show that system responses can
 - 122-

be improved with integral action and that appropriate design of the inner loop

can eliminate the oscillations.

Obviously in a heat conduction problem, another difficulty in designing a

control scheme is the lag time between a control action and the response of the

outputs. For the heat conduction process, this lag leads to large overshoot

when the feedback loop has high gains but will not lead to 1nstab1l1ty as long as

the measurements are placed within the first and last third of the system. If the

measurements are very near the edges of the system, even high gains will not

lead to overshoot; whereas, high gains leads to system instability if the measure-

ments are within the center third of the system. Thus it would seem appropri-

ate to add some sort of anticipatory control. i.e., add some derivative action.

This should reduce oscillations and keep the system stable up to higher gain.

None of the common multivariable control design techniques incorporates the

use of derivative action, since it is often difficult to physically incorporate into

the process and can lead to difficulties for step input changes and for noisy sys-

tems. Nevertheless, the application of derivative action is considered for the

(0.4,0.6) case.

Additionally, the inner-~oop decoupling control strategy was considered.

Theoretically, it led to a perfectly non-interacting system for several measure-

ment locations, but the design was based on the third-order model and could be

useless for the exact system. However, simulations show that it is an excellent

control scheme for the heat conduction process, even when the measurement

locations are not those specified by the technique. With inner-loop decoupling,

it was shown that

 - 123-

0
(2.7-6)
0

with the measurement locations (0.2,0.8) or (0.6,0.4) and arbitrary f 11 and f 21 .

Using additional proportional control with gain ~ in the ith loop, the response

to a set point change is

(2.7-7)

where

k,A.;.
c. = l+kiA.
1
A1 = C12-f- (2.7-8)
11

At.= 1
f21(2-2~c12)

f 11
Consider the first loop. The offset is obViously 1.0 - C1 = f k . Thus no offset
u+ 1

occurs if f 11 = 0 or as k 1 -+co. However if f 11 is negative, y( t) is unstable. 1 The

simulations verify these results. The application of integral action to eliminate

the offset is also considered.

Finally, perfect non-interacting compensation and steady state decoupling

are considered. The latter is obViously undesirable since it leads to dependent

control. Additionally, perfect compensation sacrifices closed-loop performance

to non-interaction and thus leads to poor dynamic behaVior. Simulations using

the perfect, non-mteracting compensator led to numerical dimculties.

Simulations of the closed-loop responses were performed for step changes

with various measurement locations. For the system with symmetric outputs

1. Note that c 12 is positive for the cases of interest.

 - 124-

and control action, both outputs respond identically to a set point change of

(1,1), 2 and the responses for (0,1) are reversed from that for (1.0). Simulations

were performed using both the third- and tenth-order process models. 9

2.7.1 Case I: llewru.rements z1 = 0.2, Z2 = 0.8

Previous analysis showed that this case is stable and diagonally dominant.

Thus high gains should reduce offset, interaction, and response time. Figure

2.7-1 shows the third-order response to a step change (0,1) with feedback gains

1.0 and 50.0. Although interaction is small at a gain of 50.0, such gain is quite

high, and steady state offset is still about 5%. 4

This case was one shown to be applicable to inner-loop decoupling. Using

extra measurements and an appropriate inner-loop compensator, the third-

order model was shown to reduce to a completely non-interacting system for

which classical methods could be used to complete the design. Third-order

simulations (Figure 2.7-2) verify the results predicted from the model. Interac-

tion is completely eliminated and the response is non-oscillatory. Additionally

with f 11 =f =0, proportional control does not result in steady state offset.
21 As

f 11 and f 21 increase, offset increases, and as the gain increases, offset decreases.

Negative values for f 11 and f 21 lead to instability.

Even the tenth-order model acts qualitatively as expected. Simulations

(Figure 2.7-:3) show that high gain leads to overshoot and eventually to instabil-

ity. Nevertheless for moderate gain. there is only a very small amount of

interaction. Thus even though the gain is somewhat limited, offset can be elim-

inated and interactions greatly reduced while still having excellent response
2. This notation means that the set points are Yd1 = 1.0 and Yd8 = 1.0 fort~ 0 with Yc11 =Yc1 11 =0.0
fort< 0.
3. Note that the tenth-order model is taken to represent the exact process.
4. In comparing the simulations, note that the time axes vary s:ignitl.cantly.
 - 125-

time. For example. with a gain of 20.0, the response time is about 0.08 and

there is negligible interaction. With simple proportional control. even with a

gain of 50.0, the response time is 50% more, while interaction and offset are

significant. Finally, integral action can be added but is not necessary since

f 11 =f =0 leads to
21 a response with no offset. Also as shown in Figure 2.7-3,

offset increases if f 11 and f:H are greater than zero.

2.7.2 Case U: Mee.surements Zt = 0.3, Za =0.7

This case is similar to the previous ln that high gains should reduce

response times. offset, and interaction, without leading to instability. The simu-

lations (Figures 2.7-4 and 2.7-5) for proportional control behave as expected.

For low gains, considerable interaction and offset arc present along with long

response times. For high gains, interaction is reduced and the system becom~s

oscillatory. However even for very high gains, the overshoot is only about 10%

with a large decay ratio, and thus the response times are short. Additionally,

the simulations show that the qualitative behavior of the responses for the

tenth-order system are identical to those for the third-order model, though

some qualitative differences are apparent at high gains.

Figure 2.7-6 shows the responses of the system with the addition of the

compensator that was designed using the characteristic locus method. At a gain

of 1.0. the third- and tenth-order responses are similar (only the tenth-order

responses are shown). The addition of the compensator has eliminated offset

and steady state interactions while adding some overshoot. However dynamic

interactions are still significant for t < 3.2. Increasing the gain reduces the

response times and interaction but increases the overshoot. At the extremely

high gain of 200.0, the tenth-order model (not shown} shows that the system is
 - 126-

unstable, although the third-order model shows a good response. Nevertheless.

short response times and little interaction can be obtained at much lower gain.

The C.L.M. procedure predicted a dramatic improvement with the addition

of an inner loop. Third- and tenth-order simulations were performed using the

design obtained in Section 2.4. The responses for the system with an inner loop

and with only proportional control in the outer loop showed considerable offset

even for step changes of ( 1.1). However, the responses (Figure 2.7-7) for the sys-

tem with the inner-loop compensator FA and the designed outer-loop compensa-

tor Gc given by Equation 2.6-20 show tremendous improvement. Although the

( 0,1) and ( 1,0) step changes no longer give identical results, the responses are so

similar that only one is shown. For gains below 8.0, both the third- and tenth-

order models show no oscillation (again only the tenth-order responses are

shown). The third-order model also shows an elimination of not only the offset

but almost all the interaction. Even though the actual system is unstable for

very high gain, the inner-loop design is excellent, since even at a relatively low

overall gain of 20.0, 5 oscillaLions and interaction are minor and response times

are short.

2.7.3 Case ill: lleesu.rements z1 =0.4, z2 =0.6

All of the design techniques showed that this case becomes unstable at high

gain (above ""'32.0). The responses (Figure 2.7-8) verify these expectations. For

low gain. the system has high interaction, large offset and long response times.

As the gain increases, the oscillations increase while the response times and

offset decrease. Nevertheless, interactions remain high at short times-t < 0.5.

Slight differences between the third- and tenth-order models are apparent at

5. The overall gain equals outer-loop ga.in-2.5-times the inner-loop gain -8.0.
 - 127-

higher gain. Although the third-order system is unstable at a gain of 35.0, the

actual system remains stable up to a higher gain (""45.0).

Both the I.N.A. technique and C.L.M. showed that the only compensator

capable of insuring system stability at high gain is a dependent controller. How-

ever as previously discussed, the use of such a controller has serious disadvan-

tages. For a set point change of (0,1) or for unsymmetric dependent control

with ( 1.1). the closed-loop behavior shows very high offset (Figure 2. 7-9), and for

symmetric dependent control, a set point change of ( 1, 1) has no effect on the

system--the output remains unchanged.

Even by squaring down extra measurements through an inner loop, stability

cannot be assured at high gain. However with the inner- and outer-loop con-

trollers that were designed using the C.L.M., the responses (Figure 2.7-10) at

moderate gain have no offset. but at the expense of increased oscillation and

response time. High overall gain(> 40.0) does indeed lead to instability.

Thus a good design has not been found for the heat conduction system with

measurements at (0.4,0.6). Obviously good feedback information is not available

due to excessive transportation lag. Thus a PD or PID controller may be useful.

If a slight amount of derivative action is added, both the third-order model at a

gain of 35.0 and the actual system at a gain of 50.0 are stable (Figure 2.7-11).

The system has a very short response time but shows some slight offset. By

further introduction of some integral action (PID control), the offset can be

eliminated. At higher gains, it becomes necessary to increase the derivative

action to insure stability; however, high gains and increased derivative action

cause numerical difficulties due to a small downward oscillation in the initial

response (t < 0.2). Since such a downward "blip" is not physically explainable,

it must be a consequence of the model reduction. This is proven in Figure 2.7-

 - 128 ~

12. The region t < 0.064 is enlarged, and it is obvious that the "blip" gets

smaller as the order of the model is increased. Finally, Figure 2.7-12 shows the

response with derivative action with the 'blip" artificially removed. This

response is physically reasonable due to the transportation lag between the

inputs and the outputs. However, although this technique allows higher gain,

derivative action is still not advisable due to problems with implementation and

process disturbances.

Finally the case (0.6,0.4) was shown to be applicable to inner-loop decou~

pling. Thus with the measurements at (0.4,0.6), a permutation matrix could be

used to convert the system to that necessary for use of this technique. How-

ever, the tenth~order response for the system with the permutation matrix and

inner~loop decoupling shows instability at moderate gain due to the excessive

transportation lag with the outputs at a distance of 0.6 from the inputs. The

inner-loop structure designed for the (0.6,0.4) case is applied directly to the

(0.4,0.6) case. Surprisingly the results (Figure 2.7-13) are excellent! Although

the interactions are not completely eliminated, they are small and may be

reduced further by improving the outer-loop design with the C.L.M.. The choice

f 11 =f 21 =0 eliminates offset, and even moderate gain leads to little interaction

and small response time. Unfortunately, the system is still unstable at high

gains.

2.? .4 Cue IV: JIElBSUl'eiilents z1 =0.4, Z2 =0.8

This case gives further insight into the system since the outputs are not

symmetric and since the root-loci have vertical asymptotes for the third-order

model. Due to the lack of symmetry, simulations are considered for the set

point changes (1,1), (1,0), and (0,1). As expected, the responses of the two out~
 - 129-

puts are not identical, and the measurement at z2 = 0.8 responds faster than
that at z 1 = 0.4. Although the third- and tenth-order models show similar
responses for low gain. they differ considerably at higher gains. The actual case

shows much worse behavior than the third-order model predicts. For propor-
tional control (Figure 2.7-14), the offset is reduced as the gain increases. How-
ever even at a gain of 50.0. the responses are quite bad. The actual system

shows a high degree of oscillation and a long response time. The output at z2 =
0.8 has an overshoot of 30%, while the other has an overshoot of 80% for a set
point change (1,1). The effects or interaction are clearly illustrated by the set
point change (0,1). The measurement at z2 , whose set point was changed, has a
fast response while the other output, whose set point remained unchanged
shows significant oscillation and long response time. Ir instead the set point
change ( 1,0) is considered, the measurement at Z:a is only slightly affected. Obvi-
ously the output at z2 = 0.8 is much more stable than that at z1.

For a gain of 100.0, the third-order response is stable and shows shorter
response times. However for all of the set point changes, the response of the
output at z 1 is unstable while that at Z:a remains stable. Even when only the set
point at z2 is changed, the . interactions cause the other output to become

unstable. The instability of the tenth-order model can be explained through a

root-locus analysis, since the tenth-order model has additional poles and zeros
which can cause the vertical asymptotes to curve into the right half plane.

Since high gains are desired to reduce offset and interaction and since very

high gains lead to an unstable response at z1 , it seems best to impose high gains
in the feedback loop on z2 and moderate gains on the output at z 1 . Figure 2.7-15
shows the responses with gains 25.0 and 100.0 imposed on the two outputs. The

output at z2 responds quickly and is only slightly affected by interaction, while

the other responds more slowly.
 - 130-

The precompensator that was designed using the C.L.M. is then added to the

system. The C.L.M. showed that the compensator should reduce interaction and

lead to a stable system with high accuracy and integrity. The responses

(Figure 2. 7-16) show that though otJset has been eliminated, dynamic interac-

tion is only slightLy reduced. Again it is eVident that the output at z 1 has too

much transportation lag. Its reponse is highly oscillatory and becomes unstable

at very high gains, even though the third-order model predicts stability. Thus

although the C.L.M. leads to a compensator that reduces interaction, the insta-

bility problem is still present. As in the (0.4,0.6) case, some derivative action
may be useful if very high gains are desirable. Finally even if an inner loop is

added, the C.L.M. showed that little improvement is obtainable. This is verified

using system simulations.

 • 131 -

.•

~
0+-........
1
~........~........r -........~....~~--~....--~........~
0. oc o. 50 l. 00 1. 50 2. 00 2. so 3. 00 3. 50 ....oo

(a)

....
0

...
0

l
]a~------~~====~~~==~~~~==~~==~
0

...
0

0+-....--~........~----r-----r---~~--~----~~----~
'o. 00 0. 04 11. 01 II. l2 11. l & II. 20 0. 24 II. H 11. 32

(b)
Figure 2.7-1
Third-Order Simulations with Outputs z1=0.2, z2 =0.8
Set Point Change (0,1)

a. Gain= 1.0 b. Gain =50.0

 • 132-

0
0

.
!
i
.E
I!

!1
!!
1I

0
"'.;
'o.oo 0. so 1. 00 1. SO 2.00 2.SO 3.00 3.50 •• oo
llonll11nd TIM

(a)

0
0

"'
0

"'_;

~"
.-
~
!
0
0

..".
t

I
... 0
~
l
!!

!
'
~
0
"'
0

"'d
'o. oo 0. 08 o. 16 o. 24 o. 32 0. 40 0.46 0.~6 0.64
•-Hzttl T1•

(b)

Figure 2.7-2
Third-Order Simulations with Outputs z 1 =0.2, z2 =0.B
Inner-Loop Decouplin.g with f 11 = f 21 = 0
Set Point Change (0,1)

a. Feedback Gain = 1.0 b. Feedback Gain= 20.0

 - 133-

..

.
!
•
l
e
i
...I
'!
..
ii
0

.,.0
.;
'o. oo o. 40 o.eo 1. 20 1. 60 2.00 2. 40 2.80 3.20
...,...11%14 Tl•

(a)

""
0

"'
0

"'
!
!!.
&

.:t"
! .
~
li

0
"'c
'o. oo o. 04 o. 08 0.12 o. 16 o. 20 0.24 0.28 0.,2
llo,...l1zl4 H•

(b)

Figure 2.7-3
Tenth-Order Simulations with Outputs z 1=0.2. z2=D.8
Inner-Loop Decoupling
Set Point Change (0,1)

fu =fa!l =0: a. Gain= 1.0 b. Gain= 20.0 c. Gain= 100.0

fu = f21 = 1: d. Gain= 20.0
 - 134-

.
g

g
.,;

!• 8
i ...
i
.j -

0
u

iO.oo 0.02 0.06 o.oe 0.10 0.12 0.14 o. 16

(c)

"'...;"'

"'..:
!• 0

i 0
..:
r:
a
I 0

...... 0"'
•
-=
li 0
0
d

0
"'ci
'o. oo o. 04 0.08 0. 12 0. 16 0.20 0.24 0. 28 0. 32

(d)

Figure 2.7-3 Continued

 • 135-

-:::
1j
1. so 2.00 2. so 3.00 '·so •• oo
..
0

ciI
llo,.lln<l Tl•

(a)

..
0

.....

.j
c
!... ------ -~--------

t
.a..
..I
11
!!
'ij
o.oe D. 12 0.16 o. 20 D. 2• 0.211 o. ~2
Nol'llllized n.
::
9

(b)

Figure 2.7~4
Third-Order Simulations with Outputs z 1=0.3, z2 =0.7

Set Point Change (1,1): a. Gain= 1.0 b. Gain= 20.0

Set Point Change (0,1): c. Gain= 1.0 d. Gain= 20.0
 - 136-

..
a

.
N

a
....
<>-+-.---....,,----....,..----,----,-------,-------·-·------.,-·-·
'o. oo o. so t. oo t. so 2. oo 2. sa s. aa

(c)

.
a

!~ ---------~--------- ------------~---
a
i :
.
~
!:
!
1.. g

...
a

0+-------r-----r-----,----r---~-----,-----~--~
'a. 00 0. OS 0.10 o. IS 0. 20 0. 2$ 0. SO 0. !IS G. 40

(d)

Figure 2. 7-4 Continued

 - 137-

<::>

-
"'

.
0

Q
I
Norw11zed Tl•

(a)

...:
0

...<>

~
1 ."'ci
....t
i~
... 0

~
ii 0
0

~ 00 o. 04 o. 08 0.12 o. 16 o. 20 D. 24 0.211 0.3:1

.
0

c:iI
No,...11ztd T1•

(b)

Figure 2.7-5
Tenth-Order Simulations with Outputs z1 =0.31 z2 =0.7

Set Point Change (1~1): a. Gain= 1.0 b. Gain = 20.0

Set Point Change (0 1): 1 c. Gain= 1.0 d. Gain = 20.0
 - 138-

....:
0

.. 0

!
~
j .. ------------------------------------
Q
0

t
11
t
!
0
•ci
~
ii
--------------------------------
...
0

~~--~~--~~~--~~--~~--~~----~----~----~
..'o.oo o.so 1.00 1.so 2.oo 2.so :s.oo s.so •.oo

(c)

"'

0
IN
..:

...
!
;;
0

ell "'c)
t
:
~

...
0

...i 0
:;
ii Cl
Cl
c)

...
Cl

0
'o. oo o. 0' o. 011 o. 12 o. 16 0.20 0.26 o. 28 o. 32

(d)

Figure 2.7-5 Continued

 - 139-

""""
N

0
"'

•• oo ••• o 5.60 I .• a

0
"'9

(a)

.
0

0.20 0,25 0.30 0.35 0.48

florwlfud Tfooe

(b)

Figure 2. 7-6
Tenth-Order Simulations with Outputs z 1 =0.3, z2 =0.7
C.L.M. Compensation Defined by Equation (2.6-10)

Set Point Change (1,1): a. Gain= 1.0 b. Gain= 20.0

Set Point Change (0,1): c. Gain = 1.0 d. Gain = 20.0
 - 140-

....

..
"'0+------r----~~----~----~------~----~----~----~
'o. oo o. aa 1. so 2. to 3. 20 •· oo •· to s. &o &, 111

(c)

.."'

...
0

di~----~------~----~----~------~-----r----~r---~
1
0. 00 0. OS 0. 10 0. 15 O. 20 0. 2S 0. 30 0. 35 0. tO

(d)

Figure 2.7-6 Continued

 • 141-

g
N

.....

2.00 2.50 3.00 3. so •• 00

.....
9

(a)

..
.....
..
....

0.10 o. 20 D. 30 o. 40 o. 50 0.60 0.70 o.ao

..
;"'

(b)

Figure 2.7-7
Tenth-Order Simulations with Outputs z 1=0.3, z2=0.?
Outer-Loop C.L.M. Compensator Defined by Equation (2.6-20)
Inner-Loop Compensator FA
Set Point Change (1.1): a. Overall Gain= 1.0 b. Overall Gain= 20.0
Set Point Change (0,1): c. Overall Gain= 1.0 d. Overall Gain= 20.0
 - 142-

0
0
..;

0
"'
!
~

! ----------------------------
i
j~
!
i g
a &~--==-~-~-~-~-~-~-~-==-~-~-:=-=-~-~-----------------------

o.so 1. DO !. so 2. 00 2. 50 :s.oo :s.so •. oo

(c)

.."'
N

"'
..:

.
c
~
;;
/!1.

..t
:>
I!

!
$
iI
0
"'
9o.oo D.IO 0.20 o. :so G. tO o.so 0.&0 0.?0 0.10

(d)

Figure 2. 7-7 Continued

 -143-

Cl
•

..
Cl

..
0

0•+-----r---~----~----~----~----~----~--~
'o. oo o. so l. oo 1. so 2. oo 2. so !1. oo !1. so '· oo

(a)
.."'
."

..
0

c~----r----,----~-----r----~----~----~--~
'o. 00 0. 50 1. 00 1. SO ~. 00 2. SCI 3. DO :S. 50 l. 00

(b)

Figure 2.7-8
Simulations with Outputs zl=0.4, z2=0.6
Set Point Change (0.1)

Gain;;::: 1.00: a. Third-Order b. Tenth-Order

Gain= 20.0: c. Third-Order d. Tenth-Order
Gain= 35.0: e. Third-Order f. Tenth-Order
 - 144-

0
•

-------------------------------

----------------------------
.
0

0;+------~----~r-----~----~------~----~----~----~
'o. oo o. to o. 20 o. :so o. •o o. so o.ao o. 10 o. eo

(c)

0
•..:

...
0

..:
!"
:,.
& .;
..0

z:
!..
I ..
0

.;
1
~

ii

...
0

.;:~-----~------~-----------T------r-----~----~----~
'o. 00 D. 011 0. II 0. 2t D. 32 0. tO 0. tl 0. 56 0. &•

(d)

Figure 2. 7-B Continued

 ~ 145-

..
0
•

i=
j..:
I

8
~+------r------r-----~-----r------r-----~----~------,
1
0. 00 0. 01 0. II 0. 24 0. 32 0. 40 0. 411 0. 58 0.14

(e)

g
..;

'~'o. oo 0. 011 o. 16 o. 24 0. 32 0. S6 0.14

(f)

Figure 2.7-8 Continued

 • 146-

"'"'
-
I
~ 0
i.. "'0

It g
... "'
~
i:
i 0 I

.."'
o.so 1. 00 1. so 2.00 2.50 3.00 3.50 t.OO

Figure 2.7-9
Tenth-Order Simulations with Outputs z 1 =0.4, z2 =0.6
Dependent Control Action Defined by Equation (2.5-Bc)
Set Point Change (0,1)
 - 147-

g
,.;

Cl
II>

o.so t.oo 1. so 2.00 2.50 3.00 3.50 •• 00

(a)
g
..;

~
:;
!
1 - --------------~---------------------
1
l Cl

Cl
II>

c+----~------T------~-----~---~------~-----~------
'o.oo o.so 1.00 t.so 2.00 2.so s.oo s.so •. oo

(b)

Figure 2.7-10
Tenth-Order Simulations with Outputs z 1=0.4, Zr;>=0.6
Outer-Loo'P C.L.:M. Compensation Defined by Equation (2.6-21)
Inner-l.Dop Compensator FA
Overall Gain= 20.0

a. Set Point Change (1.1) b. Set Point Change (0,1)

 - 148-

"'.;.

0
0
.;

...
!
0

i "'_;

0.08 0. 12 a. 1s 0.20 0.2~ o. 28. 0.32

! llol'lllliZ.d Tl•
il 0
"'

0
0

-?

(a)

0
0
..;

o. 32 a. 40 o. 48 0.!>5 a. 64'

"'.....
9

(b)

Figure 2.7-11
Simulations with Outputs z 1 =0.4, z2 =0.6
PD Control Action (To= 0.01)

a. Third-Order, Gain = 3b.O b. Tenth-Order, Gain= 50.0

 - 149-

0
1.0
0

0.32 0.40 0. 48 o. 56 o. 64
•10''
Norwllzed Tt.e

Q
0

(a)

0
CJ

"'0

0. 32 0. 40 0. 46 o. 55 0. 64
•10'' No""'H:ed Tl•

0
0

(b)

Figure 2.7-12
Initial Response Characteristics with Outputs z1=0.4, z2 =0.6
PD Control Action (To= 0.01)
Set Point Change ( 1.1), Gain = 35.0
a. 3rd-Order b. 10th-Order c. 50th-Order
d. 3rd-Order Simulation with Initial Negative Response Removed
 - 150-

0
0

0
"'ci

0. 32 0. 40 o. •e o. ss o. u
•10·1
~~o .....H ..d n ..

0
0

'

(c)

J
; ~
.i
j

li "b~+---~-.----~------~----~------r-----~----~----~
00 0. 04 0. 08 0. 12 0. 16 o. 20 0. 2. o. 28 o. 32
:! llol"'lll ht<S n.
i s:
I 7
g
_;
'

(d)
Figure 2.'7-12 Continued
 - 151-

"'...;
C)

c
"'
..:

.
J
i "'..:
C)

I•
! "'"'
l
Q

,
i
I. 00 I. so 2.00 2.50 3.CO 3.50 4.00

"'"'Q
I

(a)

0
Q
,;

"'
! .
• Q

j ~
!
~~
... 0

0.08 a. 12 0. 16 o. 20 0.28 0.32

0
"'9

(b)

Figure 2.7-13
Tenth-Order Simulations with Outputs z 1 =0.4, z2 =0.6
Inner-Loop Decoupling with f 11 = f 21 = 0

Set Point Change (1,1): a. Gain = 1.0 b. Gain = 20.0

Set Point Change (0,1): c. Gain= 1.0 d. Gain= 20.0
 - 152-

0
0
,;

0
""..:

.
i
I
::;
0
0
..:
t
;s

..i "' 0

0
1
ia 0
0
0

"'c:i
0. 00 I. co l. so 2. 00 :1. !>:J :;. 00 ~.00
.,.ltzef T1•

(c)

0
0
,;

"'...:

0
"'
0~-----,------,-----~~----,-----~-------r------~~--~
'o. oo o. oe o. JB o..'14 o. 32 o. ~o o. 48 o. ~6 o. &4

(d)

Fi.gure 2.7-13 Continued

 - 153-

o. 56 0•••

.
""
9

(a)

......;

0.20 o. 24 0. 21 o. S2
llo-Hzod n ..

(b)

Figure 2.7-14
Tenth-Order Simulations with Outputs z 1=0.4, z2 =0.8

Set Point Change (1.1): a. Gain = 50.0 b. Gain = 100.0

Set Point Change (0,1): c. Gain= 50.0 d. Gain= 100.0
Set Point Change (1,0): e. Gain= 50.0 f. Gain= 100.0
 - 154-

..
•

----------------------------------
!e
! ..

.....
0'+------r------~----~----~------r-----~----~----~
'o. oo o. oe o. 1s o. u o.,. o. to o. ta o. ss o. u
...... lUH n ..

(c)

.
"'..;.

0
0
..;

g
~i+------T------~----~----~------~-----r----~~-----
10. 00 O. Ot 0. 08 0. 12 O. 18 0. 20 0. 2. 0. 28 O. 32

(d)

Figure 2.7-14 Continued

 - 155-

...
...:...
I
';
:.,.
& 0
. C)

(!
:
I... ..
•.
....

!
iI

...
0 .

90.00 0. 011 0.16 o. 2' o. 32 o. '0 o." 0. 56 o. e'

(e)

.
Cl

"'

Cl
C)

.,;

~
..!..
i
t
I... 0
C)
Cl

~
iI. .."'
0

.
0
.,;
'o. oo o. o• 0.011 0.12 o. 111 "0.20 o. 2' 0.28 0. '2
llo,..liZH TIM

(f)

Figure 2.7-14 Continued

 - 156.

Cl
Cl

o. 20 0.28

"'
....
ciI

(a)

Figure 2.7-15
Tenth-Order Simulations with Outputs z 1;;;;;0.4, ze;;;;;O.B
Gains 25.0 and 100.0 Imposed on the Outputs z1 and z2 , Respectively

a. Set Point Change (1,1)

b. Set Point Change (0,1) c. Set Point Change (1,0)
 - 157-

..
0
..,

V>
N

"'
!...
• 0
j "'
e
a
i~
... 0
---------------------
'I
:!

1 0
0
0

"'....
0
'o.oo o. 04 o. oa o. 12 0.16 0.20 o. 24 0.28 o. 32

(b)

0
C>

"'....
0+-------~---------~---------~---------~------~----------r-------T--------,
'o. oo o. o• o. oa o. 12 o. 16 o. 20 o. u o. 21 o. :s2

(c)

Figure 2.7-15 Continued

 - 158-

..
g

a.se a.s•
....
ciI
llo-ltztd Tt•

(a)

...
..
"'.:
!
• 0
j :
i
i:;:
.. 0

~
1 ..
I ~
"b 0 0.10 o. 20 0.30 0.40 o.so o.so o. "10 0,110

..
0

o:j
llo-11 zed Tl•

(b)

Figure 2.7-16
Tenth-Order Simulations with Outputs z1 =0.4, z2 =0.8
C.L.M. Compensation in Outer Loop, Overall Gain= 50.0

Set Point Change (1,1): a. No Inner Loop b. Inner Loop FA

Set Point Change (1,0): c. No Inner Loop d. Inner Loop FA
Set Point Change (0,1): e. No Inner Loop f. Inner Loop FA
 - 159-

g
N

...
0

0+------r----~~----~----~------r-----~----~----~
'o. oo o. oe o. te o. 24 o. 32 o. 40 o. 4e o. sa o. &4

(c)

..
.,;

0
V>

0'+-----~------~----~----~------~-----r----~----~
'a. aa o. 1o o. 20 o. :so o. 40 o. so o. 10 o. 10 o. eo

(d)

Figure 2.7-16 Continued

 - 160-

..."'
CJ

."'
.
!
j -

...,
9·+-----~-----r-----r----~----~----~----~~--~
0. DO 0. 011 0, 18 0. 24 0. 32 0, tO 0. t8 0. 56 0. It

(e)

c
"'
0+-----~----~----~-----r----~~--~----~----~
'o. oo o. 1o o. 20 o. :so o. •o o. so o. &o o. 70 o. eo
..._111M Tloot

(f)

Figure 2.7-16 Continued

 - 161-

2.8 DISCUSSION OF CC1NTBOL ANALYSIS

Although much of the analysis and design in this project is an application

of published results, the work provides an insight into the current state of mul-

tivariable, feedback control theory. Unlike a review paper that compares the

theoretical bases of the design techniques, this project performed a complete

analysis of a distributed control problem using various currently available

methods. In particular, an extensive study of time-domain analysis, frequency-

domain design, and non-interacting control has been conducted for a two-input,

heat conduction system. Although heat conduction systems have previously

been studied in relation to optimal and modal control. the multiple-input prob-

lem has been relatively neglected.

Additionnlly, the role of the number of measurements and their location

was studied. Due to the ease of taking temperature measurements for the heat

conduction problem, outputs can be obtained as needed by the control system.

Thus the complete system state could easily be approximated using many meas-

urements and optimal smoothing or fewer measurements and state estimation.

However, since such flexibility is not available in many practical, distributed sys-

tems, the control scheme must work on a limited amount of output information.

Thus the analysis of the heat conduction system was performed with a finite

number of measurements, leading to the problems of measurement placement

and feedback loop interaction and instability. Since most design techniques

require feedback of outputs equal in number to the inputs, analysis of the use

of 'extra' measurements for improving system response, by reducing interaction

or moving system poles, was performed. This work led to the new technique of

inner-loop decoupling, which shows excellent results for the highly-interacting,

highly-symmetric, heat conduction system.

 - 162-

Although the problem that was studied is a relatively simple heat conduc-

tion process with two heat-fiux controls, the results lead to conclusions that can

be extended to general. multivariable control theory. The choice of the particu-

lar system is actually quite good since many difficulties arise in the feedback

design. due to excessive transportation lags and high interaction between the

outputs. In particular. the classical means of steady state decoupling and per-

fect, non-interacting control are not useful for this system, and thus more com-

plicated design methods are necessary. Therefore, though this project leads to

results specific to the heat conduction problem. the major contribution of the
work is in the general area of feedback control theory.

The study clearly showed that, although excessive model lumping and

reduction can lead to significant inaccuracies, a high degree of model reduction

is usually needed so that the various techniques of control design can be con-

sidered. Most methods require a low-order, state-space representation ot the

process for practical and efficient control structure design. Although the exact

lumping techniques of modal analysis and Laplace transform were easily appli-

cable to the heat conduction system, many other efficient techniques are avail-

able and are necessary for processes described by more complicated differential

equations. These methods include space discretization techniques and the

method of weighted residuals. along with other pseudo-modal procedures.

Before attempting any control design strategy, an accurate system model is

required. However regardless of the accuracy of the model. its usefulness is lim-

ited unless accurate model reduction can be performed. For the heat conduc-

tion process, modal lumping led to simple model reduction by directly specifying

the system eigenvalues. Since they were shown to increase rapidly. only the first

tew dominant modes needed to be retained to accurately represent the system.

The simulations showed that the qualitative behavior of the third-order model is
 - 163-

similar to that of the actual system, but the differences are magnified at high

gain and often lead to quite different behavior.

The project also allows for an excellent comparison of the currently avail-

able design techniques. Overall, the best technique that was studied in this proj-

ect is the characteristic locus method. Both with and without an inner loop, this

scheme provides a systematic, computer-aided design strategy in terms of high

and low frequency compensation that leads to an excellent, proportional-

integral controller. Actually at the specific frequencies used in the design. the

technique determines the best possible controller to insure stabiUty, reduce

interaction, and increase integity and accuracy, by calculating the compensator

that best aligns the real and complex frames. The method is based on manipu-

lating the characteristic loci and characteristic directions of the system to meet

the necessary objectives. Once the appropriate computer programs and plotting

facilities are available, the final design can readily be obtained by considering

various high and low frequencies at which the alignment is applied.

The inverse Nyquist array analysts can also easily be performed with the

appropriate computer facilities but does not lead to as good a control design.

The procedure involves maJ:ting the system stable and diagonally dominant.

Then a high amount of proportional gain can be used to reduce interactions.

However such high gain is often impractical and leads to excessive oscillations.

Additionally due to the lack of integral action. otfset is a significant problem at

low gain and can only be reduced by increasing the gain substantially. The

major advantage of this technique over simple multivariable root-loci analysis is

that, although both can be used to check stability, the inverse Nyquist method

also insures diagonal dominance. However, the characteristic locus anal,Y5is

shows that diagonal dominance is not a necessary condition1 for reduction of

1. However it is sufficient.
 - 164-

interaction at high gain.

Non-interaction control methods were also considered for the heat conduc-

tion system. Although perfect, non-interacting control and steady state decou-

pling have little use for this particular system, they may in general be applica-

ble. Whereas non-interacting control usually leads to an excessively complicated

controller and requires an accurate system model, steady state decoupling is

usually very simple and an excellent technique to perform in conjunction with

other control schemes. Furthermore, the inner-loop decoupling strategy, that

uses extra measurements to eliminate or reduce the interactions, seems to be

quite promising. Though further analysis of the method for a more complicated

system is necessary, it seems to be quite insensitive to model inaccuracies and

leads to a very simple feedback system with little interaction.

Other conclusions pertaining to general multivartable, feedback control can

be made. In particular, it was shown that none of the currently available design

techniques allow for derivative action. Though such control action may be use-
ful in many processes where transportation lags are signitlcant, it is impractical

to physically incorporate into the system and leads to problems with step

changes and noisy processes by requiring excessive control action.

Additionally, the project showed that the use of extra measurements may

drastically improve the control design or may have little or no effect, depending

on the placement of the commanded outputs. The extra outputs can be used

for inner-loop decoupling or for adjusting the poles of the process so that an

improved characteristic locus design can be obtained. In cases where extra out-

puts are available or easily accessible, they should be considered and may sim-

plify the control structure significantly. However, since a simple inner loop can-

not move system zeros, stability problems due to unstable finite zeros cannot
 - 165-

easily be eliminated with extra measurements. Unfortunately, no good tech-

nique is currently available to insure stability of such systems.

Finally the role of measurement location has been considered. Each choice

of measurement location leads to a completely different optimal design and

significantly affects the usefulness of each design technique, since the locations
determine the extent of interaction and system symmetry. AB expected, prob-

lems are minimized with measurements near the edges of the system, since
interaction and transportation lags are reduced.

Thus though the two-control, heat conduction process has been analyzed in
detail, further analysis could lead to additional insight. In particular, the trans-

portation lag problem should be further studied, especially in regard to current

work on computer-aided, multivariable control design for systems with time

delays (Ogunnaike and Ray, 1979). More importantly. an analysis of the stochas-

tic problem needs to be performed, along with an analysis of heat losses and
other disturbances. Finally, several computer-aided design packages should be

tried to further verify the results and spatial-discretization lumping procedures

and adaptive control techniques should be considered.

 - 166-

Chapter 3

:MATHEllATICAL MODEl. OF A

PACKED BED CATALYTIC REACTOR

 ~ 167-

3.1 INTRODUC'nON

The central role played by dynamic and steady state models in the design

and optimization of chemical processes and in the development and application

of control strategies justifies considerable effort in their development. The area

of packed bed reactor modeling has been the emphasis of a considerable

amount of research effort during the past twenty years and has really been at

the forefront of modeling research since the early 1970's With the acceptance of

new mathematical techniques for the solution of the systems of partial

differential equations encountered throughout the chemical industry. Still one

of the most challenging problems is that of packed bed catalytic reactor model-

ing.

Packed bed catalytic reactors are extensively used for carrying out exo-

thermic, gas-phase reactions. The complexities of the simultaneous heat and

mass transfer processes in such reactors have led to considerable effort in their

mathematical modeling. A major concern has been in the amount of detail

necessary for accurate description of the dynamic and steady state behavior of

these renctors (Heiberg et al., 1971). Although it is obvious that insufficient

detail can lead to a model in_capable of accurately representing the physical sys-

tem, model compleXity has often been limited by computational considerations.

Many authors have concluded that, although the mathematical treatment of

every aspect of the reactor system may be intellectually fulfilling, a model based

on such detail could be impractical since the resulting system of partial

differential equations would be computationally intractable. For these reasons,

much work has been directed at determining the processes that are of minor

importance and can safely be neglected in the models.

Such work has led to the extens1ve use of pseudo-homogeneous models--

 - 168-

those that do not distinguish between the conditions within the ftuid and those

on the solid catalyst. For highly exothermic reactions, a pseudo-homogeneous

description is often inaccurate due to the large temperature gradients that

exist between the solid and tluid phases. A more detailed two-phase model, in

which the exchange of energy and mass between the two phases is explicitly

described, is then necessary. Furthermore, in order to limit the complexity of

the heterogeneous model, radial concentration and temperature gradients, axial

dispersion and the variation of physical properties are frequently neglected

(Carberry, 1976; Finlayson. 1971; Heiberg et al., 1971; Jutan et al.. 1977).

Advances over the past decade in computational techniques for the solution

of partial differential equations-the orthogonal collocation method, in

particular-have made extensive simplifications of packed bed reactor models

unnecessary. Thm; the formulation and solution of accurate dynamic models of

chemical packed bed reactors is now possible, allowing

accurate description of dynamic and steady state reactor behavior

for process optimization, design and safety considerations,

investigation of reactor start-up or the effects of system distur-

bances, and

the analysis and design of control structures to stabilize the reac-

tor under various disturbances or to provide optimal system

recovery from input changes.

As a basis for a concerted effort in multivarible control system design, the

present study provides a mathematical modeling analysis of packed bed cata-

lytic reactors that significantly extends previous studies in the qetail of the

mathematical model and systematic consideration of all aspects of the model

development and the reduction to a state-space control representation. This

 - 169-

analysis is not intended to be specific to any particular packed bed system,

although our experimental methanation reactor is used throughout most of the

discussions, but rather to present a detailed study of modeling techniques.

assumptions, and solutions and to develop a unified approach to dynamic reac-

tor modeling. We feel that the modeling approach and the conclusions concern-

ing the model development and the importance of model simplifications

presented in this thesis should carry over to other similar catalytic packed bed
systems. 1

In particular, this thesis provides a complete modeling analysis of a packed

bed chemical reactor based on currently available computational procedures.

Various common assumptions, model structures, and numerical solution tech-

niques are discussed. Although some simplifying model assumptions are con-

sidered. their necessity and effect on the resulting system simulations are

rigorously analyzed.

After providing a cursory review of packed bed reactor modeling. the for-

mulation and numerical solution of a dynamic model, incorporating all of the

mechanisms necessary for an accurate description of the physical and chemical

phenomena occurring in industrial reactors, is presented for a packed bed reac-

tor in Section 3.3. The model accounts for axial and radial dispersion of mass

and energy. for mole changes that occur along the bed due to reaction, and for

temperature, pressure and mole dependencies of gas velocity, density, average

molecular weight. heat capacity, reaction rate constants, and heats of reaction.

Additionally, a central axial thermal well is included in the study to provide an

accurate representation of many industrial reactors, where the well is often

1. In line with this reasoning, all computer programs developed in this work for the reactor model
are modular and allow :for simple modification to study various packed bed reactors with varying
degrees of model complexity. Furthermore, the analysis presented is for the general packed bed
system with a cooling jacket and axial thermal well. The model and computer programs are set
up for consideration of simpler systems. such as adiabatic reactors and those without a thermal
well. These are merely subsets of the more general case.
 - 170-

used to obtain the temperature measurements necessary for process control.

Finally, the model is based on a three-dimensional--time, axial. and radial--

heterogeneous analysis and incorporates the effects of axial pressure gradients.

The model can then be used to study the steady state and dynamic

behavior of the methanation reactor, the effects of reactor operating conditions,

and the overall effects of various common modeling simplifications. The gen-

erality of the analysis also allows for studies of similar systems under adiabatic

operation and without axial thermal wells. These extended analyses are

presented in Section 3.4.

Unfortunately, the numerical solution procedures used throughout the

early portions of the analysis may be inappropriate for reactors with very steep

axial concentration or temperature gradients due to numerical difficulties

inherent in the model discretization. Section 3.5 considers such conditions and

presents an extended numerical solution procedure using orthogonal colloca-

tion on finite elements that remains stable even under the worst of these simu-

lations.

The analysis of the mathematical relationships describing the chemical and

physical processes within the reactor and the numerical approximation

methods leads to a computational technique for simulating the steady state and

dynamic behavior of the packed bed reactor. However, computing facilities gen-

erally available for on-line control cannot perform the necessary calculations

rapidly enough for practical control applications with the full, nonlinear model.

Furthermore, solution times for dynamic simulations with this model even make

detailed parameter studies and process optimization impractical. Thus a

simplified lower-order model is desired for on-line multivariable control and for

process studies and is developed in Section 3.6.

 ~ 171-

Finally, a major difficulty in accurate packed bed modeling is that of exces-

sive model dimensionality. Incorporation of all of the physical information

available into an accurate description of the reactor can lead to numerical

models of very high dimensionality. An analysis of the effect of the model

discretization on model dimensionality and techniques for accurate reduction of

the size of the model are presented in Section 3.7.

Before control studies can be performed using the model developed in this

work and even before extensive simulations are used for system design and

optimization, parameter estimation for the system of interest is necessary.

Although many of the parameters needed for the mathematical description of

the reactor system can be calculated directly from physical considerations, the

reaction and heat transfer parameters must be measured directly for the exper-

imental system. This step, outlined in Section 4.2, is by no means trivial and

may require considerable experimentation to obtain kinetic data, including con-

sideration of transient kinetics, and to obtain heat transfer and energy and

mass dispersion parameters.

 - 172-

3.2 REVIEW OF PACKIID BED REACTOR IIODEUNG

The area of packed bed reactor modeling was the emphasis of considerable

amount of research effort into the early 1970's. Many specific aspects of the

models were investigated, and well defined techniques for packed bed reactor

models were developed (Carberry, 1976; Froment, 1972; Hlavacek, 1970; Karanth

and Hughes, 1974a; Paris and Stevens, 1970; Smith, 1970).

Conclusions based on careful analyses of each specific aspect of the reactor

design led to modeling simplifications that were necessary due to limitations of

the available numerical solution techniques and computational equipment. The

use of pseudo-homogeneous models, the elimination of dispersion effects, and

the assumption of constant physical properties were often necessary and

became standard modeling practice. Steady state analyses dominated the

modeling efforts, since computational techniques for dynamic simulations were

not well-developed. However, these steady state models were able to provide

fairly accurate results for most investigations.

Advances over the past decade in.computational techniques for the solution

of partial differential equations such as those describing a packed bed catalytic

reactor have made extensive simplifications of the analysis unnecessary. In par-

ticular, a drastic reduction in computer time has been achieved by application

of the orthogonal collocation method, which is ideally suited to boundary value

problems of the type encountered in catalytic reactor modeling. In the colloca-

tion procedure, the solution is approximated by a series of known functions with

unknown coefficients, which are then determined by satisfying the differential

equations at a number of collocation points. As developed by Villadsen and

Stewart (1967), the technique uses orthogonal polynomials as the expansion

functions and the roots of these polynomials as the collocation points. The
 - 173-

method has been applied to various problems including the non-symmetric axial

diffusion for a tubular reactor (Fan et al., 1971), the steady state solution for a

nonadiabatic packed bed reactor (Finlayson, 1971), the simulation of a

simplified adiabatic packed bed reactor (Karanth and Hughes, 1974b), and the

modeling of a nonadiabatic packed bed reactor using a pseudo-homogeneous

approach (Jutan et al., 1977).

Obviously, we cannot attempt to provide a detailed review of packed bed

reactor modeling in this thesis since entire books or major portions of reaction

engineering textbooks (Carberry, 1976; Karanth and Hughes, 1974a; Smith,

1970) arc devoted to this subject. Instead, we wish simply to outline some of the

major advances in this area and to show the wide disparity in modeling tech-

niques and thus the importance of a unifying study such as that presented here.

Table 3.2-1 shows a brief summary of several published nonisothermal, nonadia-

batic packed bed reactor models developed during the past twenty years. This

table is not intended to present all of the models used throughout this period or

even present the major modeling techniques, but instead to describe represen-

tative models that exemplify the progress of this field during the period. The

table shows the continuous increase in model complexity due to computational

improvements and the generality of the present work in comparison to previous

studies.
 Solution Dispersion
Reference System Heterogeneowr Dyua.mic
Technique Thermal Mass
Orthogonal Axial Axial
Current Methanation Yes Yes
Collocation Radial Radial
Butane Orthogonal
Jutan et al. (1977} No Yes Radial Radial
Hydrogenolysis Collocation
Valstar et al. (1975} Vinyl Acetate Finite
No No Radial Radial
Synthesis Differences
Hoiberg et al. ( 1971) No {s.s.)
H2 + ~ Yes (dynamic) Yes Radial Radial
Finite Radial
De Wasch and Froment (1971} A+B ... C Yes No Radial I
Differences ....--2
I

Orthogonal !
..p.
Finlayson ( 1971) No No Radial Radial
Collocation
A_. B Finite
Hlavacek ( 1970) No No Radial Radial
Differences
Naphthalene Finite
Carberry and White (1969) No No Radial Radial
Oxidation Differences
A-. B-. C Finite
Carberry and Wendel (1983} No No Axial Axial
Quasi-Adiabatic Differences

Table 3.2-1
Summary of Nonisothermal. Nonadiabatic Packed Bed Modeling
 - 175-

Carberry and Wendel ( 1963) considered the influence of inter-intraphase

transport phenomena yield for consecutive reactions in an adiabatic packed bed

catalytic reactor. A simple digital computer model was developed with simple

extensions to the nonisothermal, nonadiabatic case in the absence of radial gra-

dients (quasi-adiabatic analysis). This study was extended by Carberry and

White ( 1969) to the steady state modeling of a packed bed catalytic reactor for

the highly exothermic oxidation of naphthalene over V2 05 . Their numerical

simulations demonstrate the necessity of a two-dimensional, axial and radial,

description of the temperature distribution and the adequacy of a one-

dimensional mass continuity description. Detailed computations further

showed the existence of significant interphase concentration and temperature

differences, even at steady state, for the highly exothermic reaction.

Hlavacek ( 1970) provided a unified review of the design of packed catalytic

reactors, including the formulation of modeling equations governing heat and

mass transfer in packed bed reactors and considerations of heal transfer

simplifications in packed beds and numerical solution methods for the resulting

set of nonlinear partial differential equations. Hlavacek describes both one- and

two-dimensional modeling of a packed bed reactor for a simple A ~ B reaction.

As with most packed bed analyses, he assumes a homogeneous reaction system,

in contradiction to the real heterogeneous structure of the packed bed. His

resulting system of nonlinear elliptic partial differential equations is reduced to

a set of parabolic equations by omitting the effects of axial mixing. Finally,

steady state solution of the equations is performed using finite ditierences. The

major contribution of his work is really in unifying the approach to packed bed

reactor design by showing the necessity and use of accurate reactor models in

studying design considerations.

 ~ 176-

Major advancements in packed bed reactor modeling were published in

1971. Finlayson ( 1971) presented the first orthogonal collocation solution for

packed bed reactor analysis. Although he showed the method to be much faster

and more accurate than finite difference calculations and easily applicable to

two-dimensional models with both radial temperature and concentration gra-

dients, the finite difference technique remained the generally accepted pro-

cedure for packed bed reactor model solution until about 1977 when the

analysis by Jutan et al. ( 1977) of a complex butane hydrogenolysis reactor

showed the real potential of the collocation procedure.

Also in 1971, De Wasch and Froment (1971) and Hoiberg et al. (1971) pub-

lished the first two-dimensional packed bed reactor models that distinguished

between conditions in the ftuid and on the solid. The basic emphasis of the work

by De Wasch and Froment (1971) was the comparison of simple homogeneous

and heterogeneous solutions and the relationships between 'lumped' heat

transfer parameters (wall heat transfer coefficient and thermal conductivity)

and the 'effective' parameters in the gas and solid phases. Hoiberg et al. (1971)

presented the first detailed, two-dimensional, heterogeneous dynamic modeling

analysis (a homogeneous analysis was used for steady state calculations). This

work considered the amount of detail necessary in dynamic models through a

comparison of experimental and calculated results for a packed bed reactor in

which the highly exothermic reaction between hydrogen and oxygen occurred on

a platinum catalyst. The major limitation of this work was the amount of detail

possible for numerical solution using the finite difference solution scheme.

The work by Valstar et al. (1975) provided the first real experimental com-

parisons between two-dimensional packed bed calculations and radial measure-

ments. Although their model was somewhat simple, especially for as late as

1975 (four years after the detailed modeling work by Heiberg et al. (1971) was
 - 177-

published), their experimental radial temperature measurements provided the

basis for the generally used assumption of quadratic radial temperature

profiles.

Finally, a significant advance in packed bed reactor modeling and control

model development was published by Jutan et al. (1977). They used the orthogo-

nal collocation discretization technique to reduce an accurate dynamic three-

dimensional--Lime, axial. and radial-partial differential equation model for the

multiple butane hydrogenolysis reactions in a packed bed catalytic reactor to a

state-space representation suitable for on-line control studies.

Other recent work by MacGregor and Wong (1978) and Wright and Schryer

( 1 978) deviated from the mechanistic approach taken by most studies where

models are developed by properly including the major phenomena occurring

within the system through the application of the basic physical and chemical

laws governing such systems 1'hey considered the use of statistical methods to

identify process transfer functions from empirical input! output process data.

The advantages of this statistical approach include rapid implementation

without the necessity of any specific knowledge of the procel'ls. However. these

empirical models are only valid within a very narrow region about the operating

conditions for which they have been derived and are only practical in cases

where very low-order models will suffice. With such complex systems as packed

bed reactors, such low-order models may be insufficient to accurately describe

the dynamic behavior of the process. Thus, these models cannot in general be

safely used for process optimization and design or the investigation of start-up

procedures and the effects of major system disturbances.

The current work presented in this thesis uses these past studies as a basis

to develop a unified general approach to packed bed reactor modeling and con-
 ~ 178-

trol model development. This approach includes detailed heterogeneous

dynamic modeling with complete axial and radial, mass and temperature con-

siderations and gas and solid property variations with minimum a priori

simplifications.
 - 179-

3.3 IIODEL DEVELOPIIENT FOR PACKED BED CATALYTIC REACTORS

3.3.1 Reactor System

Due to the need for careful. systematic studies of the model development

and the design and implementation of control processes, a fully automated,

completely instrumented, non-adiabatic, tubular, catalytic reactor was designed

and built (Strand, 1984). The reactor system was constructed, not specifically

for any particular reaction mechanism, but rather lo include various heat and

mass transfer mechanisms of interest in control studies and is versatile enough

for multivariable computer control studies of a variety of catalytic reactions.

Figure 3.3-1 shows a detailed schematic of the experimental system. The pri-

mary reaction process chosen for the initial studies is the methanation reac-
tion, discussed in Section 3.3.2.

Because of the difficulttes in heat removal rrom a packed bed, exothermic,

catalytic reactor and the subsequent problems with temperature and concen-

tration control. various reactor designs have been used to permit easier heat

removal. A common approach is Lhal of using multiple adiabatic reactor beds

with interstage quench cooling. In such systems, careful control of the tempera-

ture and concentration is necessary to assure that the adiabatic temperature

rise across the bed is not too large. Control of such a system was studied by

Foss et al. (1960), Silva (1976), Silva et al. (1979), and Wallman et al. (1979).

Another approach is to remove the heat of reaction through the reactor

walls into an outer jacket filled with a cooling fiuid. This approach is especially

useful for highly exothermic systems, since heat is removed continuously along

the reactor bed, but requires small reactor diameters due to ·radial heat

transfer limitations. Nonadiabatic reactors for highly exothermic systems are

 - 180-

then typically built with a large number of tubes in a cooling oil shell. Our reac-

tor is designed to simulate one of these tubes. Due to the high radial thermal

gradients in such tubes, detailed radial modeling is necessary.

The cooling system for the experimental reactor consists of a high tempera-

ture oil (Dowtherm) circulating between the reactor jacket and condenser by

natural convection. Continuous temperature measurements of the oil are avail-

able, and its boiling temperature can be controlled by adjusting the pressure of

nitrogen within the condenser. At normal operating conditions, the cooling fluid

will boil 1 in the jacket leading to a countercurrent flow of the oil through the

outer jacket. and the reactor wall temperature will be nearly independent of

length along the reactor. The coolant system was sized based on expected heat

loads of the reactor. A computer-controlled immersion heater is located in the

Dowtherm reservoir so that the Dowtherm can also be used to heat up the reac-

tor during start-up, since low operating temperatures can lead to undesired side

reactions.

The experimental system is designed for both feed-effluent heat exchange

and cold gas recycle that can lead to steady state multiplicities and instabilities

and are of great interest in control system design. The cold gas recycle can

actually be used to damp the highly exothermic methanation reaction by dilut·

ing the feed concentration, inhibiting the forward reaction by the introduction

of methane and by increasing the heat capacity of the feed mixture due to the

large specific heat of the methane product. A double-acting. reciprocating

piston-type compressor is used for the recycle, requiring an upstream water

knockout drum to remove the condensing water in the reactor effluent.

In addlLion Lo Lhe feed-effluent heat exchanger, a.n electric ·preheater is

1. The normal boiJ.:in8 point of l)owtberm is 257° C.

 - 181-

available for controlling the inlet gas temperature to the reactor. Thus the con-

trol configuration can consider the complete control of the inlet feed tempera-

ture or simply feed-effluent heat exchange.

The actual reactor bed consists of a 1.194 em radius stainless steel tube,

through the center of which runs a 0.159 em radius thermal well containing

thermocouples at various axial positions2 The reactor chamber is about 30 em

long and packed with finely ground~ nickel on alumina catalyst particles.

Because of the cooling jacket, radially mounted thermal wells are impractical. so
all internal temperature measurements are made within the thermal well, as is

common in many industrial reactors.

All feed gases to the reactor are supplied by standard gas cylinders. Each

stream is controlled by a mass fiow controller and is equipped with a regulator,

an oxygen absorber I catalyst, a filter I drier. a moisture indicator and a solenoid

shutoff valve. Under normal operations for the methanation process, nitrogen,
hydrogen and carbon monoxide are used as the inlet streams.

The computer system used for measurements and control is a Digital Equip-

ment (DEC) LSI 11123 to which all control valves, thermocouples, heaters, and

solenoid valves are connected. Concentration measurements are obtained

periodically using a gas chromatograph with a thermal conductivity deleclor.

An on-line CO I C02 detector will be installed for future control studies.

This experimental system is advantageous for multivariable computer con-

trol studies due to its flexibility, the large number of available of measurements,

feed-effluent heat exchange. the possibility of product recycle. and the inherent

time delays. The reactor can actually be used to study a large number of

processes simply by changing the inlet gases and the catalyst.

2. The thermal well is designed to contain up to 24 thermocouples.
3. 0.8- 1.0 mm average diameter
 - 182-

3.3.2 Reaction Kinetics

Alhough a large variety of reaction processes could be studied in this exper-

imental packed bed reactor, the methanation reaction was chosen for the initial

studies since methanation is highly exothermic and the temperature and con-

centration control of exothermic packed bed systems has traditionally been

difficult. since methanation is a reaction of great industrial importance and

since methanation has been well-characterized in the literature. Methanation is

one of a more general class of Fischer-Tropsch processes where carbon monox-

ide and carbon dioxide are hydrogenated to form various light hydrocarbons

and water. When using specific selective catalysts. the resulting product is pri-
marily methane. The catalyst used in our studies is Girdler catalyst G-65, a

nickel on alumina catalyst used industrially to specifically promote methanation

without the excessive formation of other hydrocarbons. Another advantage in

using methanation for the initial control studies is that by simply replacing the

catalyst, higher-order Fischer-Tropsch processes with additional control

difficulties can be studied.

Although the primary reaction on this catalyst is the methanation of car-

bon monoxide, appreciable side reactions can occur in the methanation system.

ThP.se are shown in Table 3.3-1 and include carbon dioxide methanation. steam-

shift, carbon deposition, nickel carbonyl formation and other Fischer-Tropsch

processes. By operating at H2 :CO ratios of about 3:1 and temperatures above

zooa C. most side reactions are suppressed and only the CO methanation, C02

methanation, and steam-shift reactions should be significant. Of these three

reactions, only two are independent. In light of the discussion below on C02

methanation, the CO methanation and steam-shift reactions are .taken as the

independent reactions. These reactions have been studied extensively and

ldnetic information is available.

 - 183-

CO Methanation: CO + 3~= CH + H204

C02 Methanation: C~ + 4~ = CH + 8H20 4

Steam-Shirt: CO + H2 0 = C02 + H2
Carbon Deposition: 2CO _. C02 + C

Nickel Carbonyl Formation: Ni + 4CO ,... Ni(CD) 4

Fischer·Tropsche: nCO + 2nH 2 = (CH2)n + nH20

Table 3.3-1
Reactions in Methanation Systems
(Source: Strand, 1984)

A general rate expression for CO methanation over a nickel catalyst is given

by Lee (1973) and Vatcha {1976). They report that a Langmuir-Hinshelwood

equation of the torm

(3.3-1)

where (3.3-2)

is superior to simple power law relationships. These reaction kinetics (Equations

3.3-1 and 3.3-2) are based on experimental data taken by Lee (1973) in a cata-

lytic CSTR under conditions of ideal mixing for CO methanation over a Harshaw

Ni-0101T nickel-Kieselguhr catalyst and extensions by Vatcha (1976) to include

the factor ( 1-v) to reduce the rate to zero at equilibrium. The range of validity

for this expression is given in Table 3.3-2, and empirical values for the rate con-

stants are given in Table 3.3-3. Vatcha reports that the low activation energy

indicates that the above rate expression describes the global rate, thus incor-

porating any mass transfer limitations. Gas phase concentrations can then be

used in the reaction analysis. This use of gas phase concentrations in the
 - 184-

analysis does not limit the generality of the model, since the mass transfer

effects are simply included in the rate expressions and since the interparticle

mass transfer limitations are often minimal for packed bed reactors where the

gases flow rapidly over solid catalyst particles (Jutan et al., 1977).

As far as side reactions, Vatcha (1976) concludes that

'the presence of CO strongly inhibits the methanation of

C02, but C~ barely influences the methanation of CO in
their mixtures. Thus, in a reactor C02 would remain uncon-
verted until the CO became depleted to a very low concen-
tration (typically 200 ppm on nickel) and only then would
the C02 begin to get methanated."

Total Pressure 1-69 atm

Temperature 547-755° K
F'eed Gas Composition
~:CO > 2.85
H20 < 5%
c~ < 20%
N2 < 50%
HaS < 0.5ppm
Table 3.3-2
Validity Range of Methanation Rate Expression
(Source:Vatcha, 1976)

Jlethanation Steam-Shift

mole CO mole C0 2
kou 0.075 kos 17.816
sec g cat atm L 5 atm2 g cat sec
EaM 6944.
cal
Eas 18900. cal
g-mole g-mole
K1 1.47 atm- 1 fl 0.83 atm
K2 0.73 atm-1 rl! 0.17

Table 3.3-3
Empirical Constants for Equations (3.3-1) and (3.3-3)

· Although Vatcha also shows that literature surveys on the steam-shift reac-

tion are inconclusive, we have included this side reaction in our modeling
 - 185-

analysis. This allows consideration of multiple reactions in the control model

development. A complete analysis of the steam-shift reaction is provided by Moe

(1962). Based on correlations of experimental data, he reports a rate expression

of the form

(3.3-3)

which should be appropriate over a wide range of operating conditions. The

pressure dependence of the steam-shift reaction rate

(3.3-4)

is also derived from empirical results (Moe, 1962).

Finally, the equilibrium constants Kpll and KPs are taken as functions of

temperature as given by van't Hoff's equation:

ln I<p., = KP11t + ~~~

ln~ = ~1 + ~2
The necessity of these relationships and values for the parameters in these

expressions are presented in Section 3.4.

The empirical rate expressions given in Equations (3.3-1) and (3.3-3) were

used for initial simulations. Actual rate expressions for our specific catalyst,

reactor bed and expected operating conditions were determined in preliminary

experiments using a kinetics reactor (Strand, 1984) bullt specifically for these

studies. Preliminary results show that the kinetics for our catalyst are actually

much different than those predicted by the above expressions. The dependence

of the rate on the specific concentrations of carbon monoxide, hydrogen, and

methane are significantly different, and the activation energy is much higher. In
 • 1B6-

particular. the reaction was much more sensitive to operating conditions and

especially to the temperature, leading to much steeper gradients in the axial

concentration and temperature profiles. These new methanation kinetics are of

the form (Strand, 1984)

(3.3-6)

With the rate constants shown in Table 3.3-4. The final term, e-(lt, in the rate

expression is an empirical catalyst deactivation term. The steam-shift reaction

was found to be negligible with the nickel on alumina catalyst under the prelim-
inary experimental conditions.

1.2 BxlOt4 mole CO

sec gr cat atm2
37000_ .s&_
mole
110. atm- 1
2.32 atm- 1
0.3 hr- 1

Table 3.3·4
Rate Constants !or Methanatton Reaction
Kinetics Given By Equation (3.3-6)

This rate expression differs significantly from that proposed by Vatcha

( 1976). The new expression does not include any terms explicitly for the equili-

brium (i.e., a term like 1-v in Vatcha's expression). This term was not found to

be needed under the limited planned operating conditions (relatively Low tem-

peratures and concentrations), since at these conditions the equilibrium con-

stant KPu is approximately 109 and the reaction therefore goes to nearly 100%

completion. Furthermore. experimental results showed a very high activation

energy and rapid deactivation of the catalyst. The rate expression includes an

empirical deactivation term, although the activity remains nearly constant dur-
 - 187-

~ most control experiments of reasonable duration (3~5 hours) after an initial

deactivation period of about 100 hours.

Significant insight into the mathematical modeling of packed bed reactors

can then be obtained by completely modeling the experimental reactor using

both reaction rate expressions for methanation. Because of the preliminary

nature of the new kinetic expression (Equation 3.3-6), much of the analysis

presented in this thesis is performed using the methanation kinetics expression

given by Equation (3.3-1). Conclusions in this modeling work were verified with

the new reaction kinetics, thus allowing consideration of various kinetic models

in the modeling analysis. The only major effect of the new kinetics on the

modeling work is discussed in Section 3.5.

As pointed out in this section, global reaction kinetics are used in this

analysis. These kinetics must then account for the the adsorption/desorption

on the catalyst surface and the intraparticle diffusion. However, most available

kinetic information is based on steady state data. A major concern is then the

importance of the transient behavior of the adsorption/desorption processes

and the characteristic times for the intraparticle diffusion dynamics. Although

the intraparticle diffusion can be shown to be generally insignificant in the reac-

tor dynamics:' experimental efforts are needed to gain further information on

the dynamics of the adsorption/desorption processes.

3.3.3 Formation of Complete llathematicalllodel

A two-dimensional, two-phase mathematical representation of the experi-

mental packed bed reactor is developed in this section. This mathematical

4. The time constants for the temperature and concentration profiles in the pellet to change are at
least an order of magnitude faster than the time constants for the temperature and concentra·
tion proilles in the reactor bed.
 - 168-

model significantly extends previous packed bed modeling studies through a

reduction in a priori simplifications. Detailed consideration of the common

simplifications is then possible, and the validity of certain assumptions can be

assessed in light of the combined effect of multiple assumptions and of the

overall benefits of the assumptions. The general view that modeling

simplifications will lead to a reduction in numerical solution effort and are

therefore desirable or in many cases necessary is shown to not be universally

correct with today's new computational capabilties. Furthermore, most previ-

ous modeling studies have been restricted to specific systems due to the appli-

cabiltiy of the chosen assumptions for each individual case. The approach

taken in this work was to minimize these assumptions so as to develop a general

packed bed reactor model that could be transported to other systems with

minimal efi'ort.:s

The heterogeneous packed bed reactor consisting of solid, nonspherical

catalyst particles and reacting gas is treated as two phases with the assumption

that the packed bed may be treated as a continuum insofar as changes occur
smoothly and continuously within each phase throughout the bed. This assump-

tion is generally valid for most industrial reactors and should be valid under the

conditions of this analysis (Carberry and Wendel. 1963; Hlavacek. 1970; Stewart,

1967), since the ratio of bed diameter to particle diameter for the experimental

reactor is about Z5 and the axial aspect ratio is very large (ZOO- 300). The heat

and mass fluxes can then be treated in a form analogous to Fourier and Fick

laws, respectively.

Previous investigations have often assumed that the difference between the

catalyst and gas temperatures are negligible in tubular reactors for fast flowing
5. Although this current modeling is for a general, nonadiabatic packed bed reactor with an axial
thermal well, the analysis easily extends to the consideration of adiabatic reactors and those
without a thermal well. These are merely subsets of the more general case.
 -189-

gas-solid systems. Industrial experience and experimental studies (Froment,

1974; Gould, 1969; Hoiberg et al., 1971; Jutan et al., 1977) have verified that

there is essentially no temperature dift'erence ( < 5a K) between the catalyst and

gas at steady state. Furthermore, this assumption has also often been defended

by the argument that it is difficult in practice to measure separately the gas and
solid temperatures. However, for control studies, the two-phase representation

is necessary since considerable (> 10° K) differences can exist between gas and

solid temperatures during dynamic conditions, and for control applications, the
dynamic situation is of major importance. The measurement problems can and

often are in practice reduced by measuring temperatures within an internal

thermal well rather than in either the gas or catalyst. Modeling of the thermal

well in the system is then necessary to perform accurate dynamic control.

A variety of assumptions is often made concerning the axial and radial

dispersion of mass and energy. These assumptions are based on a significant

amount of supporting literature and for most systems involve neglecting axial

and radial mass diffusion and axial energy diffusion, depending on the aspect

ratios of the reactor bed. Although these assumptions have often been neces-
sary in the past due to limitations of the computational techniques available for

numerical solutions of the model and in particular the second derivative

ditiusion terms, current techniques do not require such assumptions and may

indeed be hindered by them.

The present analysis begins by incorporating all axial and radial dispersion

effects, including axial conduction within the thermal well. Axial conduction in

the outer wall is neglected based on the first assumption presented below and

on Bonvin's (1980) results that the conduction in the outer wall is most likely

insignificant and can be neglected from the model it axial dispersion in the bed

is retained. Although many dynamic analyses of packed bed reactors have

 - 190-

neglected radial gradients to limit the complexity of the resulting model. this

work considers a packed bed system with extensive cooling at the reactor wall

and thus must account for the radial profiles within the bed (Jutan et al., 1977).

Radial temperature gradients are neglected within the thermal well due to its

comparitively small radius and high thermal conductivity.

Also accounted tor by the model are density, heat capacity, and molecular

weight variations due to temperature, pressure, and mole changes, along with

temperature induced variations in equilibrium constants, reaction rate con-

stants, and heats of reaction. Axial variations of the tluid velocity are
accounted for in the mathematical description using the overall mass conserva~

tion or continuity equation. These variations are the result of axial tempera-

ture changes and the change in the number of moles due to the methanation
reaction.

The major assumptions underlying the original model are:

1. The reactor wall temperature is equal to the cooling tluid temperature

and is independent of length along the reactor (Carberry and Wendel,

1963; Jutan et al., 1977; Smith, 1970). The validity of this assumption is

generally based on the very hiJ;h thermal conductivity of the reactor wall

and on the use of boiling ftuids or high convection in the outer cool.ing

shell. The experimental reactor is designed to have boiling fluid in the

cooling jacket, as is common in highly exothermic industrial reactors.

2. Gas properties are functions of temperature, pressure, and total moles as

dictated by the ideal gas law. The assumption of ideal gas behavior will be

accurate as long as the operating temperatures of the reactor are much

higher than the critical temperatures of the component species and the

pressures are relatively low. These conditions should be met at the

 - 191-

expected operating conditions of the experimental methanation reactor

and are in general valid for most gaseous reaction systems.

3. There is no radial velocity, and the axial velocity across the radius of the

packed bed is uniform. Schwartz and Smith ( 1953) found that the veloc-

ity across the diameter of a packed bed is not uniform for radial aspect

ratios (tube-to-particle diameter) less than about 30, due to the

significant effect of the increased void space near the wall where the par-

ticles are locally ordered. This result has been verified by Hoiberg et al.
(1971) for a packed bed reactor with radial aspect ratio about 50. They

considered a radial velocity runction suggested by experimental observa-

tions with a sharp peak about 15% greater than the mean fluid velocity
situated close to the wall. Simulations using their model showed results

virtually identical to those obtained with a uniform velocity profile.

Although the radial aspect ratio for our experimental reactor is under

30, 6 a uniform velocity profile was assumed. Preliminary residence time

distribution studies should be conducted on the experimental packed bed

reactor to test this assumption. Although in many cases it may be

desired to increase the radial aspect ratio (possibly by crushing the

catalyst), this may be difficult in our system due to the highly exothermic

solid-catalyzed reaction that can lead to excessive temperature rises

near the center of the bed. Carberry ( 1976) recommends reducing the

radial aspect ratio to minimize these temperature gradients. If the velo-

city profile in the experimental reactor is significantly nonuniform, the

mathematical model developed here allows predictive equations such as

those by Fabien and Stankovic ( 1979) to be easily incorporated.

6. Actually there are only about 12 particle die.meten~ between the thermal well and outer wall.
 - 192-

4. The physical properties of the solid catalyst and thermal well are taken

as constant, since the conditions within the reactor introduce only minor

vartations in these parameters, and the heats of reaction and gas heat

capacities are taken as linear functions of temperature.

5. Hlavacek ( 1970) has shown that radiation between the solid catalyst and

gas can signifl.cantly affect the temperature dynamics in packed bed sys-

tems operating in excess of 673° K. Since the system considered in this

work usually operates well below these conditions, radiation terms are
not explicitly included in the model. However, their effect can to some

degree be accounted for in the overall heat transfer coeffi.cients. 7

It should be noted that the model developed in this analysis may be much more
complex than that necessary for accurate description of the experimental

methanation reactor. The increased complexity and generality allow simple

extensions of the model to other systems and to include additional physical and
chemical processes such as radial velocity variations and the dependence of

heat transfer coefficients on physical properties as defined by heat transfer

correlations.

3.3.4 Mathematical Relationships

The analysis in this thesis centers around the actual reactor bed. Combin-

ing these results with a simple analysis of the external processes, such as the
product recycle and feed-effiuent heat exchanger. allows overall system analysis.

Figure 3.3-2 shows an expanded section of the reactor bed and defines the coor-

dinate system for the mathematical modeling.

A complete mathematical description or the reactor bed results using the

7. But not entirely since radiation effects are nonlinearly related to temperature.
 - 193-

six differential equations that describe the catalyst, gas, and thermal well tern-

peratures, CO and C02 concentrations, and gas velocity. These are the con-

tinuity equation, three energy balances, and two component mass balances. The

following equations are written in dimensional quantities and are general for

packed bed analyses. Systems without a thermal well can be modeled simply by

letting hu. ht.g. and Ro equal zero and by eliminating the thermal well energy
equation. Adiabatic analysis simply involves setting h.,... and h.,, equal to zero.

Total mass conservation (continuity):

=
Z 0 =
PgUg PgoUgo

Energy balance for the gas:

(3.3-B)

Then for the reactor bed after assuming kzg and krg constant,

(3.3-9)

BT
r=R0 kr8 7jf-= hq(T8 -Tt)
BT
-kr8 ~= h.,..8 (T1 -T..-)

z=O kq,7f:-=
8T
hsg(T -T 8 9) - UgCpfgE(To-T1 )
8T
z=L -kr.g,Tz-= hs,g(T8 -T8 )
 - 194-

Note that the convective term in the outlet boundary condition is generally

assumed negligible (Heiberg et al., 1971). This assumption is used

throughout the ensuing analysis. Also note that the gas heat capacity, cPs·

gas density, p 8 , and gas velocity, u 8 , are functions of position and time due

to their dependence on mole changes, pressure, and temperature. Original

calculations considered the gas heat capacity and thermal conductivity as

linear functions of temperature since expected changes in molar composi-

tion throughout the bed had minimal effects ( < 1%) on these properties.

However, later analysis replaced the thermal conductivities and heat

transfer coefficients with dimensionless Peclet and Biot numbers which

were taken as constant throughout the bed.

Energy balance for the catalyst:

Using a similar analysis to that for the energy balance of the gas and

after assuming constant physical properties of the solid phase,

(3.3-10)

r=Ro krs ~· =ht.s(Ts-Tt)

r~Rl -krs ~ ~ h.w.(Ts-Tw)
z=O BT8
~~az-=h88 T8 -T8
( )

-kzs ~; = hag(Ts--Tg)

The heats of reaction for the methanation and steam-shift reactions are
 - 195-

taken as linear functions of temperature:

i =S, M (3.3-11)

based on literature data and standard temperature dependence of the

heat capacities of the gas components. A verification of the necessity and

applicability of this relationship along with the specific parameters for

methanation and steam-shift is given in Section 3.4.

Energy balance for the thermal well:

Assuming constant physical properties in the thermal well,

(3.3-12)

z=O
z=L

De Wasch and Froment (1971) discuss the calculation of the wall heat

transfer coefficients for the fluid and gas phases based on a lumped wall
heat transfer coefficient. Furthermore, radial heat conduction in the ther-

mal well is neglected since it should be of minor importance for a thin solid

well.

Mass balance in the reactor section:

Since two independent reactions are expected to be important within the

reactor, a mass balance must be written for each of two independent

species. Due to the sensitivity of the concentrations of the gaseous species

in the reactor under the planned operating conditions and to difficulties in

accurate concentration measurements, these two are taken as CO and C02

in this analysis. Both of these species can accurately be detected using a

 - 196-

chromatograph with a simple thermal conductivity detector or with a. con-

tinuous CO/C02 analyzer. Note that i = 1 refers to CO and i =2 refers to

c~ in the following analyses.

The equation of continuity in cylindrical coordinates is (Bird et al ..

1960)

(3.3-13)

But

BN,e = O
1. Be

thus

If we then incorporate the void fraction and apply the continuity equation

for species i along with Fi.ck's law of diffusion and the assumption of con-

stant ditiusivity and no radial velocity

oci _ o(C(ug)
---
Bt
o [c Bx,; + -
- Bz + D21 -Bz Bz
l
a [ Bx,; Rt
Dr -
r or rc Br - e-
l
(3.3-14)
where Rt = Rw-Rs . ~ = Rs .
i:}~
-=0
Br
z=O Ug(Ci 0-c,)
ax..
= -cDzaz
a~
z=L -=0
Bz
 - 197-

The rate terms R11 and R5 are taken to be global rates, incorporating all

mass transfer limitations. This allows bulk phase concentrations to be

used throughout the analysis. Although species concentration is proper-

tional to mole fraction, 0( = CX(, complications arise since the total number

of moles decreases as the methanation reaction progresses. Therefore to

simplify the analysis technique, the mass balances are written using molec-

ular weights and mole fractions based on inlet conditions. Letting obe the
moles of CO reacted in the methanation reaction per total inlet moles,

(3.3-15)

the following relations hold

(3.3-16)

The advantage of this formulation is that the molecular weight based on

inlet conditions, Mg. is constant. Thus

(3.3-17)

Then after application of the overall continuity, Equation (3.3-15) becomes

a [rp --+
D,. -
+- 2pgJCir aoj_ ~Mg
o:Ki .....;....>-<-::-:- (3.3-18)
r ar g ar 1-20 ar £'

z=O

z=L
 - 198-

Furthermore, algebraic manipulation of the boundary conditions using the

reaction relationship 6 =X'10 +X20 -X'1 -X'2 leads to8

z=O

z=L

Additional relations:
Finally, relationships for density and pressure changes are necessary. The

ideal gas law leads to

(3.3-19)

The changes in the pressure along the bed are taken as linear by assuming

uniform packing and negligible wall effects. he overall pressure drop

across the bed is simply defined by the Ergun equation (Perry and Chilton,

1973)

6P- 1[150(1-E)J.Lg + 175 ]1-c Ug, (3.3-20)

- Dp . P&o ~ e Dp

where D = 6(volume of particle)

P external surface area of particle

This could easily be replaced with a Blake-Kozeny type relationship

P =Po at z=O (3.3-21)

if the original assumption is inadequate.

B. The stipulation x10 + X2 0 #

boundary conditions.
* is needed for l"'.igorous mathemaucal derivation of the reduced
 ~ 199-

It should be noted that the importance of the continuity equation in the

general modeling presented throughout this thesis may be questionable since its

only use is in evaluating actual velocities Within the reactor bed as intluenced by

the mole, temperature, and pressure changes. Because of the use of mass veloci-

ties (pgug) throughout much of the analyses, the importance of the actual veloci-

ties is really restricted to analyses where pressure relationships such as the

Blake-Kozeny equation or velocity effects on heat transfer parameters are con-

sidered. Since our analysis is set up to be general and allow the inclusion of
these effects, the continuity equation is necessary. As seen later, very little

increased computational effort is introduced by retaining this equation, since it

is solved as a set or algebraic equations.

3.3.5 NUIIlericu.l Soluliuu.

Before attempting to solve the system of partial differential equations. they

are reduced to dimensionless form. The axial parameters are normalized with

respect to the reactor length, L, radial parameters with respect to the outer
radius, R1 , time with respect to the characteristic time L/u80 , and the remaining

parameters with respect to the steady state inlet conditions. The concentra-

tions are actually normalized with respect to the inlet steady state concentra-

tion of CO rather than with respect to each individual component, since some

inlet concentrations (except CO) may be zero during normal experimental con-

ditions.

The resulting mathematical system consists of six coupled, three-

dimensional, nonlinear partial differential equations along with nonlinear alge-

braic boundary conditions, which must be solved to obtain the temperature

profiles in the gas, catalyst, and thermal well, the concentration profiles, and
 -200-

the velocity profile. These equations are presented in Appendix 2. In their

present form. direct solution is not possible. However, approximation tech-

niques are available to reduce the equations to a set of tl.rst-order ordinary

ditl'erential equations in the time domain. The resulting system can then be

solved numerically using a variety of standard techniques. In this work, the

method of orthogonal collocation is used for this reduction, since it has proven

to be an extremely powerful technique in reactor modeling (Bonvin, 1980; Fin·

layson. 1971; Jutan et al., 1977).

3.8.5.1 Solution Techniques

Considerable emphasis has been placed during the past twenty years on

numerical solution techniques for complex nonlinear systems of the types com-

mon in chemical reactors. Traditional finite difference schemes are being

replaced by applications of the methods of weighted residuals. Finlayson ( 1972,

1980) has presented and compared the application of these various techniques

for the nonlinear annl.ysis of problems in chemical engineering.

The finite difference method involves dividing the domain up into intervals

with the boundary points between intervals being called the grid or mesh points.

Then for a continuous function across the interval, a Taylor series expansion

can be used to deduce ditrerence formulas for first and second derivatives. If

the ditrerential equations are written at each grid point using the difference for-

mulas and the values at the first and last grid point solve the boundary condi-

tions, enough equations are available to solve for the value of the function at

each grid point and thus provide a representation of the solution. For most

chemical systems. the resulting equations are nonlinear, and accurate solution

requires a large number of grid points. Solution of the system is not trivial but
 -201-

can usually be obtained quite rapidly using various standard numerical pro-

cedures (Dahlquist and Bjerke, 1974; Davis, 1984; Finlayson, 1980).

The finite difference method can easily be extended to multidimensional

systems by applying the same techniques in each of the dimensions. However,

the scale of the numerical problem increases dramatically with the number of

dimensions. Since nonadiabatic packed bed reactors consist of two important

spalial dimensions, radial and axial, along with the lime dimension and usually

require a large number of grid points. accurate solution using the finite

difference scheme is often computationally prohibitive and may limit the com-

plexity of the mathematical model.

The method of weighted residuals presented in detail by Finlayson (1 972.

1980) is a general method of obtaining solutions to both linear and nonlinear

systems of partial differential equations and is often used in one of its forms to

reduce the computation time from that of the finite difi'erence technique. In the

method of weighted residuals, the unknown exact solutions are expanded in a

series of specified trial functions. that are chosen to satisfy the boundary condi-

tions, with unknown coefficients that arc chosen to give the 'best' solution to the

differential equations:

y(x) =~ 0 (x) + t
k~
CA:¥"A:(x) (3.3-22)

These trial functions are substituted into the differential equations, and the

result is the residual (R). This residual is weighted by functions characteristic

of the particular method, and the weighted residuals are minimized over the

domain of the independent variable (V). In particular, the weighted integrals of

the residuals are set to zero:

 -202-

j = 1. 2, ... , N (3.3-23a)

where the inner product is defined by

(3.3-23b)

The method of weighted residuals is comprised of the following basic tech-

niques, depending on the choice of the weighting function (Finlayson. 1972):

Subdomain Method

The domain is divided up into N subdomains, V;, and the weights are
chosen as

X inVj
X not in Vj

The differential equation, integrated over the subdomain, is then zero.

As N increases, the differential equation is satisfied on the average in

smaller and smaller subdomains and approaches the exact solution

everywhere.

Collocation Method

The weighting functions are chosen to be the Dirac delta function

(3.3-25)

Thus

fv Wj R dV = R l'1<1 (3.3-26)

This technique then forces the residual to be zero at N specified collo-

cation points. As N increases, the residual is zero at more and more

points and presumably approaches zero everywhere.

 -203-

Um.st. Squares Method

The least squares method uses the weighting functions

BR
W·=-- (3.3-27)
J Bcj

so that the mean square residual fvR2(e;.,x) dV is being minimized.

The mean square residual is zero for the exact solution, so that, as

the number of parameters is increased, the mean square residual

gets smaller and the approximate solution approaches the exact solu-

tion.

Galerkin's Jlethod

In this method, the weighting functions are chosen to be the trial

functions, which must be selected as members of a complete set of

functions. (A set of functions is complete if any function of a given

class can be expanded in terms of the set.) Also according to Finlay-

son (1972),

"a continuous function is zero if it is orthogonal to every

member of a complete set. Thus the Galerkin method
forces the residual to be zero by making it orthogonal to
each member of a complete set of functions (in the limit
as N-+ oo )."

Method of 'Moments

In the Method of Moments, the weighting functions are chosen as

w·J = x1-t (3.3-28)

Thus successively higher moments of the residual are required to be

zero.

All of these methods of weighted residuals have proven to be quite powerful and

have been shown by Finlayson ( 1972, 1980) to be accurate numerical techniques

 ~ 204-

superior to finite ditl'erence schemes for the solution of complex differential

equation systems.

Another potential solution technique appropriate for simulations of the

packed bed reactor is the method of characteristics. This procedure is suitable

for hyperbolic partial differential equations of the form obtained from the

energy balance for the gas and catalyst and the mass balances i! axial disper-

sion is neglected and if the radial dimension is first discretized by a technique

such as orthogonal collocation. The thermal well energy balance would still

require a numerical technique that is not limited to hyperbolic systems since

axial conduction in the well should be significant.

3.2.0.1 Orthogonal Collocation

Of the various methods of weighted residuals presented in the previous sec-

tion, the collocation method and in particular the orthogonal collocation tech-

nique described in this section has proven to be quite effective in the solution of

complex, nonlinear problems of the type typically encountered in chemical reac-

tors. The basic procedure was used by Stewart and Villadsen (1969) for the

prediction of multiple steady states in catalyst particles, by Ferguson and Fin-

layson ( 1970) for studying the transient heat and mass transfer in a catalyst

pellet, and by McGowin and Perlmutter (1971) for local stability analysis of a

nonadiabatic tubular reactor with axial mixing. Finlayson (1971, 1972, 1974)

showed the importance of the orthogonal collocation technique for packed bed

reactors.

The orthogonal collocation method has several important differences from

other reduction procedures. With other techniques, difficulties are often

encountered in deriving the values of the integrals involving complex nonlinear

 -205-

terms. In collocation, it is only necessary to evaluate the residual at the colloca-

tion points. The orthogonal collocation scheme developed by Villadsen and

Stewart ( 1967) for boundary value problems has the further advantage that the

collocation points are picked optimally and automatically so that the error

decreases much faster as the number of terms increases. The trial functions

are taken as a series of orthogonal polynomials which satisfy the boundary con-

ditions and the roots of the polynomials are taken as the collocation points.

Thus the choice of the trial functions and collocation points is no longer arbi-
trary, and further analysis (Finlayson, 1972) shows that with this choice, low-

order collocation results are more dependable. A major simpllficatlon that

arises with this method is that the solution can be derived in terms of its value
at the collocation points, instead of in terms of the coefficients in the trial func-

tions, and that at these points the solution is exact.

A rigorous description of the technique requtres appropriate definitions of

the orthogonal polynomials Pm(x) as

(3.3-29)

with degree m and order m+l. The coefficients are defined so as to require the

orthogonality condition
b
J. w(x)Pn(x)Pm(x) dx n=O, l, 2, ... , m-1 (3.3-30)

to be satisfied for weighting functions w(x) ~ 0. For boundary value problems,

the solution is expanded in terms of orthogonal polynomials with the first term

satisfying the boundary conditions followed by a series that has unknown

coefficients, with each term satisfying the homogeneous boundary conditions.

Var.ious expansions are then possible. The most common and useful are:
 -206-

(3.3-31)

YN(X) = ~ y(X-t)"(x)
i=l

Since derivatives are actually expressed in terms of the solution at all of the

grid points in the collocation scheme rather than simply in terms of the neigh-

boring grid points as in finite difference schemes, the orthogonal collocation

technique then leads to computer programs that are relatively simple and to

numerical solutions that require a very few number of collocation points in

comparison to the number of grid pointR neceRsary for a similar finite difference

solution.

3.3.6.3 Mod.el Reductwn

Finite difference methods have traditionally been used for reduction of par-

tial differential equation systems to ordinary differential equations or even to

systems of algebraic equations that can readily be solved using simple numeri-

cal techniques. However in chemical reactors, this method has a major draw-

back. For the system of six coupled, three-dimensional. nonlinear partial

differential equations describing the methanation reactor, if we assume that

even 5 grid points are adequate for discretizing the radial direction and 30 grid

points for the axial direction, 9 we have a mesh of 150 grid points. At each grid

point, there are six ordinary differential equations in time. Hence the total

number of ordinary differential equations necessary to describe the system

would be 900! Much too large for extensive simulations or for control.

. The method of orthogonal collocation on the other hand has been very sue-

9. Other studies have used up to 20 radial grid points and 10000 axial grid points!
 -207-

cessfully applied to chemical reactors and provides an accurate means of

transforming partial differential equations into a reasonably small set of ordi-

nary differential equations. This procedure is applied to the reactor model to

discretize both the radial and axial dimensions, leaving a manageable set of

ordinary differential equations in time.

3.3.5.3.1 Radial Collocation

The first step in the solution procedure is discretization in the radial direc-
tion. This involves writing the three-dimensional differential equations as a pos-

sibly enlarged set of two-dimensional equations at the radial collocation points

with the assumed profile identically satisfying the radial boundary conditions.
An examination of experimental measurements (Valstar et al., 1975) and typical

radial profiles (Finlayson, 1971) indicates that radial temperature profiles can
adequately be represented by a quadratic function of radial position. The qua-

dratic representation is preferable to one of higher order since only one interior

collocation point is then necessary, 10 thus not increasing the dimensionality of

the system. The assumed profile is of the form:

(3.3-32)

with the number n of interior radial collocation points taken as one. This profile

must satisfy the boundary conditions and must be exact at the collocation

points r =t;Do. rc• and 1.0, where the interior collocation point rc is selected as

the zero of the appropriate orthogonal polynomial. This formulation is

equivalent to the trial function

y(x)::::: b +cx+x(l--x) ~ a,;Pi-t(X)

i=l

10. Along with two boundary points, there are three collocation points.
 -208-

(3.3-33)

given byVilladsen and Stewart (1967) for nonsymmetric profiles. 11

Rigorous application of the procedure as presented by Finlayson ( 1972) or

the equivalent procedure presented below of satisfying the boundary conditions

with the assumed profile, leads to the coefficients c4(19.() for the gas and tern-

perature profiles in terms of the temperatures at the collocation point, within

the well. and at the outer wall. Since the objective of radial collocation is to

eliminate the radial derivatives, Equation (3.3-32) is then substituted into the
partial differential equations.

For example, consider the radial temperature profile in the gas. At the

three radial collocation points r = 9'o· rc, and 1.0, let the gas temperatures be

98o, air. and ®g 1 and assume that the radial profile is quadratic:

(3.3-34)

This profile must satisfy the boundary conditions

(3.3-35)

where A.trg and A.wrg are the dimensionless radial Biot numbers at the thermal

well and cooling wall. respectively. The profile must also be exact at the three

collocation points:

e8o =do+ d19'o + d29'§

air= do + dlrc + d:!r; (3.3-36)
9g1 = do + dl + ~

After rearranging and eliminating 98o and 9g 1 •

11. Due to the presence of the thermal well, a nonsymmetric function is nece83ru"Y to dei!Cribe the
profile between r = =
9'o and T 1.0.
 -209-

(3.3-37)

The expressions for c4(1).() in terms of 9~,.. 9t and 9w can be simply obtained by

applying Cramer's Rule. These solutions are shown in Appendix 3.

Since

(3.3-38)

the dimensionless form of Equation (3.3-10) becomes

(3.3-39)

where Sg and 0 8 are now the temperatures at the radial collocation point rc, and

p 8 and cp are now dimensionless parameters, normalized with respect to Lbe

11

inlet steady state values. Similar results are obtained for the energy balance of

the catalyst. Although the solutions of the d.ift'erential equations are then

obtamed for the temperatures at the radial collocation point, the temperature

at any radial point can easily be determined using the radial profile given by

Equation (3.3-34) along with the solutions to Equation (3.3-37) for d,('l).()

presented in Appendix 3.

The radial collocation point is then selected as the zero of the orthogonal

polynomial for cylindrical geometry between r ='i"o and r =1.0. To be rigorous,

the entire analysis could be based on the shifted variable

T\Do
r'=-- 0 ~ r' ~ 1 (3.3-40)
l~Do

However, lengthy algebraic manipulations show that the results are identical.
 -210-

Nevertheless, it is convenient to use r' to find the radial collocation point, which

is then the zero of the Jacobi polynomial tor a nonsymmetric system between

zero and one. 12

For radial concentration profiles, a quadratic representation may not be

adequate since application of the zero fiux boundary conditions at r =rp 0 and
r =1.0 leads to d2 = ds =0. Thus a quadratic representation for the concentra-
tion profiles reduces to the assumption of uniform radial concentrations, which

for a highly exothermic system may be significantly inaccurate. Although addi-

tional radial collocation points greatly increase the dimensionality or the result-
ing model. they may be necessary to accurately express the radial concentra-

tion protlles. Preliminary analysis in this section considers only one interior
radial concentration collocation point. although a detailed analysis of this
assumption is presented in Section 3.4.5. In that analysis, an assumed concen-

tration profile of the same form as Equation (3.3-32) is solved with several inte-

rior radial collocation points to determine the radial concentration coefficients.

The mass balances are then satisfied at each collocation point.

Thus, the original differential equations have been reduced from three-

dimensional in variables (, r, and ~to two-dimensional in variables ( and ~by

orthogonal collocation in the radial variable -r. The reduced temperature equa-

tions are a function of the dependent variables at the radial collocation point rc,

but incorporate radial information via the well and wall temperatures. Radial

profiles can be generated Cram the collocation equation using the expressions

for the coefficients di(~.() presented in Appendix 3. Similarly, the continuity

equations are functions of the concentrations at each collocation point for mul-

tipoint radial collo~::ation. With one interior collocation point for the concentra-

tion analysis, the system consists of six differential equations describing the
12. The s:in&Je radial collocation point is taken as r' = 0.5 for the methanation system.
 -211-

overall continuity, energy balance for the thermal well, energy balance for the

gas and catalyst in terms of conditions at rc, and mass balances for CO and C02.

This shows the incentive for using only one radial collocation point, since the
number of two-dimensional partial differential equations is the same as the ori-

ginal number of three-dimensional equations.

3.3.5.3.2 Axial Collocation

Since the resulting system after radial collocation is still too complex for

direct mathematical solution, the next step in the solution process is discretiza-

tion of the two-dimensional system by orthogonal collocation in the axial direc-

tion. Although elimination of the spatial derivatives by axial collocation greatly

increases the number of equations, 18 they become ordinary differential equa-

tions and are easily solved using traditional techniques. Since the position and

number of points are the only factors affecting the solution obtained by colloca-

tion, any set of linearly independent polynomials may be used as trial functions.

The Lagrangian polynomials of degree N based on the collocation points ( 1

i=O, 1, ... , N+l {3.3-41)

are used here. The ditlerential equations are then collocated at theN zeros, {,t.

k ::: 1, 2, .... N, of a Legendre polynomial. Since only one lagrangian is non-zero

at a collocation point and since the residual is set equal to zero at this paint,

the coetlicient of the Lagrangian term is equal to the solution at that point.

Following the development of Villadsen and Michelsen ( 1978), the assumed

axial profiles are:

13. Since more tha::1 one axial collocation point is generally necessary.
 -212-

j =1, 2
(3.3-42)

Since the l.,;(() are known functions based solely on the collocation points, the

differential operators can be applied a priori:

A;;. = dl.,;(() J (3.3-43)

d( ~j.

Then after operating on the assumed solutions with the differential operators
and substituting into the partial differential equations. the residuals are set

equal to zero at the collocation points.

The collocation points are calculated using programs given by Vllladsen and
Michelsen ( 1978) for calculating the zeros of an arbitrary Jacobi polynomial
plf.Bl(x) that satisfies the orthogonality relationship

(3.3-44)

where c, is a constant, t5v is the delta function and a and (3 are chosen based on

the geometry of the system. For the methanation reactor, zeros of Ptf·0 >(z) are
used as collocation points.

The resulting equations in terms of conditions at the collocation points ~

are:

Catalyst Energy Balance

(3.3-45)
 -213-

Gas Energy Balance

.(3.3-46)

:U:a.ss Balances

+ 4y~ oa~ Jz] +

( 1-2o) 2
[
~
(3.3-47)
k:;:; 2
 - 214-

where

Thermal Well Energy Balance

Overall Continuity

(3.3-49)

Note that i = 1, 2.... , N for all of the equations except the last where

i = 1, 2, ... , N+ 1, that the G.>t coefficients are the result of radial collocation and

are presented in Appendix 3, and that all other dimensionless parameters are
defined in Appendix 2. Furthermore, the reaction terms R'll and R's are calcu-

lated at the conditions at the collocation points.

The resulting equations are a set of 6N+l ordinary differential equations

along with algebraic boundary conditions. The solution procedure is further

simplified by solving the last equation for v~ as a set of algebraic equations,

using the gas temperatures and derivatives of the gas temperatures from the

solutions of the remaining differential equations. 14 Additionally, simple algebraic

manipulation allows for explicit solution of Vio, eto, etN.f.l' Yio and Y;.:NH' while

solution for @llg, @IIN+t' 9.1o, and 9gN+l requires the SimultaneoUS solution of four

coupled algebraic equations. The resulting dynamic model is then a set of 5N

coupled, nonlinear ordinary differential equations and N+5 coupled, nonlinear

algebraic equations, where N is the number of interior axial collocation points.

Typical steady state profiles along a reactor obtained by other authors

(Jutan et al., 1977; Heiberg et aL. 1971) indicate that these profiles can be

represented by relatively low-order polynomials. However, temperature profiles

with steep gradients, as may be likely in a highly exothermic system such as

methanation, may require higher-order polynomials. Although approximation

error is reduced by increasing the number of collocation points. numerical

problems with fitting higher-order polynomials to process curves may result.

Section 3.7 discusses the model dimensionality in detail. Presently, let it suffice

to say that 6 - 8 axial collocation points are generally sufficient. leading to a sys-

tern of 30 - 40 coupled, nonlinear ordinary differential equations and 11 - 13

coupled, nonlinear algebraic equations.

14. i.e., effectively substitut.iiJ8 Equation (3.3-46) into Equation (3.3--49) for d: .
de
 -216-

S. S. 5. 4 Numsrical Simulation

Steady state solutions can be obtained by carrying a dynamic simulation to

steady state or by setting the time derivatives equal to zero in the ordinary

differential equations and then solving the resulting system of 6N+5 algebraic

equations. In the latter technique, solutions for very steep axial profiles often

show numerical convergence problems. These can be reduced by using better

initial guesses or by varying the actual solution algorithm. In extreme cases. it

may be necessary to carry the dynamic solution part way to steady state and

then use its results as an initial guess of the steady state solution.

Due to the large size of the mathematical problem and the complexity of

the nonlinear equations, both dynamic and steady state solutions require

powerful algorithms for the solution of nonlinear algebraic equations and for

the solution of initial value problems in ordinary differential equations. Consid-

erable analysis of various solution algorithms led to the final selection of the

techniques described in this section due to their robustness under a great

variety of conditions and their relative speed of solution in comparison to other

techniques. Most other procedures exhibited convergence problems. Even the

selected techniques fail to converge under certain conditions. but in general are

flexible enough to allow solution with the adjustment of several convergence

parameters.

Although the solution of the algebraic equations is relatively simple for the

dynamic simulations. it is extremely difficult for steady state solutions with poor

initial estimates of the solution profiles. Two methods are used in the computer

simulation programs developed in this work for the solution of the systems of

nonlinear algebraic equations. These are based on algorithms by Powell (Rabi-

nowitz. 1970) and Brown (1967). Although Brown's algorithm is in most cases
 -217-

more powerful than that of Powell, it is in general significantly slower. Thus

although the computer programs are written to use either technique, Brown's

algorithm is only used in those cases where Powell's algorithm fails to converge

to an appropriate solution.

Consider the system of nonlinear equations

Jc =1. 2, .... n (3.3-50)

where xis the vector of unknowns (x 1 , K2· ... , xJ. F(x) is the sum of squares of

the residuals

F(x) = f:
.t=l
[ft(x)] 2 . (3.3-51)

Powell's technique tor the solution of nonlinear equations is a hybrid of the

classical Newton-Raphson procedure and the Levenberg/Marquardt steepest de-

scent method. The new algorithm retains the fast convergence of the Newton

method but ts modified to take steps along Lhe sLeepe::oL descent direction of

F(x), if it seems that the classical Newton iteration diverges.

The major difference between this hybrid algorithm and the standard

Levenberg/Marquardt iteration is that Powell's procedure does not require

explicit expressions for the derivatives of the functions, but instead uses succes-

sive values of f,;(:xU)) ( i = 1, 2, ... , n: j =1, 2, ... ) to build up a numerical approxi-

mation to the Jacobian matrix. Note that r.;) is the jth estimate of the solution.

The technique further includes several parameters that can be adjusted to

improve convergence for each particular system. One common difficulty with

this algorithm is that the method may converge to a stationary point or F(x),

although this may not be a global minimum. The theory underlying this pro-

cedure, along with examples and a sample Fortran program are given by Powell

(Rabinowitz, 1970).
 -218-

In the event that Powell's algorithm fails to obtain a solution to the equa-

tions, a different initial estimate of x can be attempted or Brown's (1967) qua-

dratically convergent algorithm can be used. Basically, Brown's procedure con-

sists of

"expanding the first equation in a Taylor series about the

starting guess, retaining only linear terms, equating to zero
and solving for only one variable, say x..:. as a linear combi-
nation of the remaining n-1 variables. In the second equa-
tion, Xt is eliminated by replacing it with its linear represen-
tation found above, and again the process of expanding
through linear terms. equating to zero and solving for one
variable in terms of the now remaining n-2 variables is per-
formed. One continues in this fashion, eliminating one vari-
able per equation, until for the nth equation, we are left
with one equation and one unknown. A single Newton step is
now performed, followed by back-substitution in the tri-
angularized linear system generated for the x.'s. A pivoting
effect is achieved by choosing for elimination at any step
that variable having a partial derivative of largest absolute
value ... (Brown. 1 967)

Dynamic solutions of the mathematical model ror the packed bed reactor

also require a powerful method for initial value problems in ordinary differential

equations. The procedure selected in th1s work 1s a standard Adams-Moulton

predictor-corrector technique. This procedure is based on the formula

(3.3-52)

where the differential equations are defined as

~= f(x,y(x)) (3.3-53)

In the Adams-Moulton case, f(x:,y(x)) is approximated by an interpolation poly-

nomial determined by the values at each iteration, f"+ 1 • fn, ... .fn--.~: +1· In the par-
Licular routines used in Lhis work, k is taken as three and the predictor-

corrector formulas are

 - 219-

y!n:F) -Yn = ~~ (55fn -59fn-1 -37fn -2-9fn-:3)

(3.3-54)
Y~Cfr) = Yn+ ~: (9fn+l+19fn-5fn-l+fn-2)

This method is combined with an automatic control of the step size and uses the

method of Runge-Kutta-Gill to start the integration process and to restart the

integration any time the interval size has been changed.

Thus the system of ordinary differential equations describing the reactor is

solved using an Adams-Moulton predictor-corrector technique with the method

of Runge-Kutta-Gill being used to start the integration process, and the non-

linear system of algebraic equations is solved using Powell's hybrid algorithm

(Rabinowitz, 1970) or Brown's ( 1967) quadratically convergent algorithm.

The computer simulation programs developed for this system use modified,

double precision versions of the Caltech library routines MODDEQ, NSESl. and

NSES2 for these algorithms. Combining these techniques with a variable time-

step analysis, based on increasing the time-steps as the derivatives decrease

during the approach to steady state. leads to an efficient solution procedure for

obtaining dynamic reactor responses. All programs are written in double preci-

sian and are modular so as to allow for easy modification or use with other reac-

tor systems by simple subroutine replacement. These computer programs and

complete operating instructions are included in Appendix 4.

Although the technique described above is useful for simulating the full,

nonlinear ordinary differential equation model, solution times can often be quile

long. even up to several hours of computation time on a Digital Equipment (DEC)

Vax 111780 for a fifteen minute simulation due to the size and nonlinearity of

the system. These ·can be reduced significantly by linearizing the set of ordinary

differential equations around the steady state solution. 15 The linearization is not
15. This of course requi.res the ability to directly solve for the steady state solution.
 -220-

lim1ted to only the rate expressions but is performed on the entire differential

equations and is complicated by the dependence of the velocities, densities, heat

capacities. and other gas properties on the temperatures and compositions

throughout the bed. A linearization around steady state conditions leads

directly to a model of the form

:i(t) =Ax(t) + Bu(t) + D (3.3-55)

where the state vector x includes the solid, gas, and thermal well temperatures,

and CO and C02 concentrations at the collocation points. The control vector u

consists of the expected disturbance and control variables, including cooling

wall temperature, inlet velocities. inlet gas temperatures, and inlet concentra-

tions of all species.

This linearized model can be solved explicitly in terms of matrix exponen-

tiation for time periods t 0 to ~ where u remains fixed:

(3.3-56)

This analytic solution can then be simulated using eigenfunction evaluation of

the matrix exponential. Further explanation of this analysis is presented in

Sections 3.6 and 3.7.

 l\ t ~ ...

j !II •

ro
ro
......

PNIDUCT~
·-...----i-1

Figure 3.3-1
Process Diagram for Experimental Catalytic Reactor
(Source: Strand, 1984}
 -222-

0
.,..._....,
I

IR 0
I I
I I
I I

FEED

+
•••••••••••••• •::::::::::::'========:::•r- -
••••••••••••
••••••••••
•••••••••••
-r -- 0
••••••••••••
·:::::::::::. z
~~~~~~~~~~~~~,.,,0~~.::~
••••••••••••
••••••••••••
••••••••••••
:::::::.·::··-·c.•..JJ::::;:;::::;:;:.;:;:;:;::<•
••••••••
•••••••
••••••••••••
••••••••••••
••••••••••••
•••••••••••
••••••••••••
••••••••••••
••••••••••••
••••••••••••
••••••••••••
••••••••••
••••••••••••
••••••••••••
•••••••••••
••••••••••••
..............
••••••••••••• - ----L

,CATALYST l:t~ ~m~1~ ~ l~ ~l~l

\THERMAL WELL

Figure 3.3-2
Expanded Section of Reactor Bed
 • 223-

3.4: IIODEL ANALYSIS

The objective of developing a complete mathematical description of an

experimental reactor is not simply intellectual fulfillment, but rather to study

the steady state and transient effects of various parameters. operating condi-

tions, and modeling assumptions on the behavior, or actually the predicted

behavior, of the experimental process. This allows gaining significant insight

into the operation of the system without unnecessary lengthy and often hazard-

ous experiments. Using the fully developed model. specific experiments can be

planned, and the experimental design can be optimized to minimize potential

hazards and to focus on the areas of concern.

The steady state and transient effects of various parameters and assump-

tions on the mathematical simulations of the packed bed catalytic reactor are

now examined using the complete mathematical model developed in the last sec-

tion. These studies lead to conclusions concerning the importance of various

physical and chemical phenomena and to the importance of a generalized

overall model structure. Note that the importance of this study is not simply in

defining our experimental system, but rather in its general applicability to

packed bed reactor modeling. All of the analyses presented in this thesis are

based on numerical solutions of the full, nonlinear model using the methanation

kinetics of Vatcha (1976), although many analytic solutions using the reduced

linear model developed in Section 3.6 and solutions with the methanation kinet-

ics proposed by Strand (1984) were used as verification of the important conclu-

sions.

Unless otherwise stated, all figures presented throughout this and the fol-

lowing sections are based on the reversible methanation kinetics of Vatcha

(1976), all axial gas and catalyst temperature profiles are at the radial colloca-
 -224-

lion point, and all concentration profiles are mole fractions based on the total

inlet moles as defined by Equation (3.3-17). Furthermore, all simulations are

based on the kinetic parameters in Table 3.3-3, along with the modeling parame-

ters and operating conditions presented in the next section. The purpose of this

is to retain consistency throughout this thesis so that the various simulations

can easily be compared.

3.4.1 Modeling Parameten

The numerical values of the parameters used for this analysis are based on

published results of other packed bed analyses. Typical values of the major

parameters are shown in Table 3.4-1. The kinetic parameters are given in Table

3.3-3. Unless stated otherwise, these are the parameters used in the prelim-

inary simulations.

Catalyst Parameters Thermal Well Parameters

Cp. 0.23 Opt 0.12

Ps 1.04 Pt 8.02
kzs,krs 0.0005 kzt 0.039

Heat Transfer Parameters Reactor Parameters

Us.g 17.02 L 30.00

Uts 0.02 Ro 0.159
Utg 0.14 R1 1.194
~s 600.00 Penu: 10.00
~ 13.00 Pernr 2.00
"-trs=~s 7.16 p~ 2.00
Xtr2 =X..,.,., 1.25 Peru- 8.00

Table 3.4-1
Typical Reactor Parameters
 -225-

The physical reactor parameters were measured for the experimental

methanation reactor. The void traction must be empirically determined for the

specific catalyst used in the experiments due to differences in catalyst crushing.

The catalyst used for the preliminary experiments is standard Girdler G-65

methanation catalyst. Its physical properties are shown in Table 3.4-2, which is

reproduced from the specification sheet from United Catalysts Inc. Although

the density of the catalyst is given in the specification sheet, the overall density

should be measured for the specific experimental conditions since the catalyst

is actually crushed substantially for our use. The heat capacity of the catalyst

is calculated over the expected temperature range of operation based on the

heat capacities of the individual components of the catalyst and the catalyst

composition. Results of this ana.lysis a.re are shown in Table 3.4-3. Since the

temperatures within the reactor bed should generally remain within 100° K, a

constant heat capacity can be used.

The Lhermo.l well properties are calculated using data for stainless steel

(type 304) from CRC Handbook of Chemistry and Physics (Weast, 1976) and

Chemical Engineers' Handbook (Perry and Chilton, 1973). The heats of reaction

and equilibrium constants for the methanation and steam-shift reactions are

also readily available. linear regression of data provided by United Catalysts

Inc. is shown in ~,igures 3.4-1 and 3.4-2 over the expected temperature range of

operation and is used to determine linear temperature relationships for the

heats of reaction and relationships for the equilibrium constants based on van't

Hoff's equations (Equation 3.3-5). Notice that the heats of reaction and the

equilibrium constants for both the mcthanation and steam-shift reactions vary

greatly over the temperature range of interest. Although the use of constant

values for these parameters would be highly inaccurate, relationships based on

van't Hoff's equation for the equilibrium constants and linear equations of the
 -226-

M.thanatlon Catalylt ~ICh~~land

Ph~lcal Prop41ftlee

C..Uiylt l)J141, C.tllyat Type ........... G-e.s EffectMII: Jaroary 25. 1110
form and lla Form .................. Tabt.ta 5u!>lt'MdiiS: January 31, 11J711

---
SIZit • . • . ...........•.. \4" X \4..

Chemical Nl" . . . . .............. 24-27 ...,o................. <0.15 ---

---
Composition Alr01 .... , · ... · · · · · · · · ~ MQO ....... : . ......... <0.10
c.o .................. 4-t
SIOr .................. <0.30 LOf to Conatant Weight at:
C (AI Graphite) ......... !S-5
s• (MlXImum) .......... 0.()!5 10000F· ............. 10.0 Mu.
Na• (Mnlmum) ........ 0.30

Physical A. Bulle O.nalty, lba./ cu. ft. .............. 85 :1:: 5

PropertiM
B. Surfac. Arwa, m'lg .................. 3:5-4!
'\'he surface arwa is obtained by a modified BET method which conalaw of
nitrogen adaorptlon bV the aampla. Tile sampla it ;round to a fine powder
(~100 IT'Ifth) and purged with nitrogen at 200•c to 1 conatant night.

C. Porw Volume, cc/g .................. 0.15-0.25

~ porw volume is obtained with a mercury porsimatar at 110.000 peig which
corrwsponda to the total pore volume or paras greater then 211.2 All;stroma
dia.,.ter.

D. Cruth Strwngth, Minimum AVIrage• ..... SIO.O lbl.

No Morw Than 5% Leu Than .......•.. -40.0 I be.
Aj)carahra- HYdraulic oms with horizontal Platas of which the bottom plat.
move• vertically. In Mrias with the bottom plata Ia a prnaurw gauge which
~ordl the prnaure axartad on 11'11 tablat

PrOCidunt- The tab lot Ia pta cad on lla aida and the lonr plat. It rwiHd to
the tablet with Z&ro pounds pressure on tile gauge. The preuurw is then
tntroasao on tilt t>onom plata until tilt tablet cru•'- and ,,... p..uurw at
breakage ia rweol'ded.
The cruah on • minimum of tnnty·fiVI tablets It obtained anclll'le IWI1'111Je It
taken arlth.,.ticalty.
To prwtrwat tne Nmpt•, Clry tne catalyat 11 400"F lor thfll houra and allow to
COOl.
Cruth St11tngth Range on
Individual Tablets ........•...•....... -40-150

•...__ _....., - - - "' Quooll!r Clonlolltl

Table 3.4-Z
Girdler G-65 Catalyst Specifications

form given in Equation (3.3-12) should be very accurate. As discussed in Section

3.3, these equations were incorporated in the original model. The coefficients

for our experimental conditions calculated using linear regression of the data

presented in Figures 3.4-1 and 3.4-2 are given in Table 3.4-4.

The remaining parameters for the mathematical description of the experi-

mental system are the heat transfer variables and the reaction kinetics.
 -227-

Temperature ( 0 K) Cp (cal/ g
1
o K)

550. 0.225
575. 0.228
600. 0.230
625. 0.231
650. 0.232
675. 0.233
700. 0.235
725. 0.237

Table 3.4-3
Temperature Dependence of Catalyst Heat Capacity

Although these need to be estimated from initial experiments for our specific

catalyst and reactor bed (described in Section 4.2), preliminary values from

literature are used. Effective thermal conductivities and heat transfer

coefficients are given by De Wasch and Froment (1971) for the solid and gas

phases in a heterogeneous packed bed model. Representative values for Peclet

numbers in a packed bed reactor are given by Carberry ( 1976) and Mears

(1976). Values for Peclet numbers from 0.5 to 200 were used throughout the

simulations.

For consistency, several sets of typical operating conditions are used for

the simulations presented in this thesis. These are shown m Table 3.4-5 and wtll

be referred to as standard Type 1, II. or Ill conditions. These three significantly

different sets of conditions were chosen to cover a wide range of potential

operating conditions. Type I corresponds to operation at moderate to high tern-

peratures. pressures. and fiowrates wtth relatively low inlet CO and H2 concen-

trations and small amounts of inlet CH4 , C0:2. and H20 either from recycle or

from the upstream process when the methanator is being used to cleanup the

process stream prior to ammonia synthesis. Type II corresponds to realistic

conditions for the industrial use of methanation in synthetic natural gas pro-
 -228-

Jlethanation Steam-Shift

~1 -29.44 KPst 4.39

KP1L2 26341. KPsa -4615.
.6Hu. -6.14 l'1Hs1 -2.44
.6Hu2 -48350 .6Hsa 10760 .

Table 3.4-4
Coefficients for Equations (3.3-5) and (3.3-12)

duction. Note that the inlet methane concentration is much higher than in Type

1. This large amount of methane significantly reduces the reaction rate. Finally,

Type III corresponds to single-pass laboratory experimental conditions where

only CO and H2 are fed to the reactor. Flowrates and temperatures are rela-

tively low.

Type I Conditions

~:=0.40 T0 =573° K Tw=573" K

Parameters:
Ug 0 = 75.0
em/sec P=20 atm
XCH =0.02 xco2 =0.03 xa 20 =0.024
Inlet Mole Fractions: 4
xco=0.03 xa,=0.20

Type II Conditions

Parameters: = 0.57
1! T0 =573" K Tw=573" K
u& = 75.0 em/sec P= 10 atm
XcH4 =0.60 Xco2 =0.015 Xa 2o =0.02
Inlet Mole Fractions:
Xco =0.06 xa,= 0.19

Type ill Conditions

t=0.57 T0 =510o K Tw= 530" K

Parameters: Ug0=14.0 em/sec P=ZO atm
XcH4 =0.00 xco2 =0.00 Xf! 2o =0.00
Inlet Mole Fractions:
xco =0.04 xHe= 0.12

Table 3.4-5
Typical Operating Conditions
 -229-

3.3.1 Stee.drState Beha<fior

Steady state axial gas and solid temperature profiles at various radial posi-

tions and concentration profiles for the standard Type I operating conditions

are displayed in Figure 3.4~3, and the corresponding radial temperature profiles

are displayed in Figure 3.4-4. AB mentioned previously, the concentration

profiles show the mole fractions of CO and C02 based on the total inlet moles.

The definition of these values is the moles of CO or C02 divided by the total inlet

number of moles. This definition is used rather than standard mole fractions in

most situations since the total number of moles decreases rapidly in the bed.

At the conditions chosen for Type I operation, the methanation reaction is

quite rapid and approaches completion. The steam-shift reaction. which is

much slower, leads to a slight formation of C02 . Note that the overall conver-

sion of the C02 is actually quite small (less than 1%).

The steady state temperature prot:lles (Figures 3.4-3b and 3.4-4) show that a

'hot spot· is present about half way through the reactor bed and is predominant

near the center of the bed. The presence of such a hot spot is a result of the

cooling jacket and is common in nonadiabatic packed bed reactors. The steady

state temperature profiles show a difference of up to 10" K between the solid and

gas temperatures, and later transient results show differences up to 20" K (lead-

ing to differences of over 20% in the reaction rates), thus providing initial

verification of the necessity of the heterogeneous analysis.

In the small region near the cooling wall. even the steady state temperature

d.i.fferences between the solid and gas are significant due to the higher heat

transfer between the solid and cooling wall than that between the gas and the

cooling wall. The radial temperature profiles (Figure 3.4-4) also verify the neces-

sity of the radial temperature analysis, since the radial gradients are significant
 -230-

throughout the bed. Notice that except near the cooling wall the catalyst tem-

perature is greater than the gas temperature due to the exothermic reaction

occurring on the catalyst surface.

The standard Type II operating conditions are much milder and result in

only about 40% CO conversion and much lower reactor temperatures due to the

lower pressure, the larger void fraction, and the large amount of methane in the

feed. Steady state axial temperature and concentration profiles for these condi-

tions are shown in Figure 3.4-5, and the steady state radial temperature profiles

are shown in Figure 3.4-6. Again the temperature differences between the solid

and gas are significant.

3.3.2 Dynamic Simulations

The real power of the model developed in this work is not in steady state

analyses, since this is relatively simple and has been investigated in detail in the

past, but rather in the transient or dynamic simulations necessary for control

design. This model has been used to simulate the effects of various process dis-

turbances and input changes. Under normal reactor operating conditions, step

or pulse changes in inlet gas temperatures, concentrations, or velocity or

changes in cooling rates can significantly affect the behavior of the process.

These disLurbances must be understood for optimal system operation and con-

trol.

Figure 3.4-7 shows the effect of a 10% drop in the inlet gas temperature

(from 573" K to 515.7" K) on the axial temperature profiles within the reactor

for standard Typ~ I operating conditions. Since such a disturbance would, in

general. not occur instantaneously, the inlet temperature was actually changed

from its initial to final value in a 0.5 second ramp. The complexities of the
 - 231-

resulting reactor profiles are enhanced by the behavior of the cooling system.

Although the inlet gas temperature is reduced, the cooling jacket temperature

remains unchanged (T., = 573" K). Thus the cooling jacket acts as a heating sys-

tem in the early part of the reactor and as a cooling system in the later part,

thus in effect transferring heat from the later stage to the early stage of the

reactor. Figure 3.4-7 also shows that although the inlet gas temperature is

reduced, the steady state outlet temperature actually increases by about 8° K.

since the hot spot shifts further down the reactor, effectively reducing the cool-

ing region. As expected, the responses of the concentration profiles (not shown)

are much faster than that of the thermal profiles. indicating the possible appli-

cability of the quasi steady state approximation for the concentrations (this will

be discussed in detail in Sec lion 3. 7).

Figure 3.4-8 shows the axial gas and solid temperature profiles during start-

up operation. Notice that the 'hot spot' in the reactor moves down the bed as

the heat generated from reaction heats up the catalyst particles. Also note the

significant temperature difference between the catalyst and gas in the early part

of the reactor where conversion is rapid. These differences are even more pro-

nounced (over zoo K) near the center of the bed and near the outer wall. 1
Figure 3.4-9 shows the temporal behavior of the catalyst, gas, and thermal

well temperatures at ( == 0.38 and ( = 1.0 for a step change in the cooling fluid
temperature from 573c K to 593° K as would occur through increasing the nitro-

gen pressure in the Dowtherm condenser. This figure exemplifies the slow

response of the catalyst and thermal well temperatures due to their high ther-

mal capacitance. As discussed later, the gas temperature merely follows the

behavior of the solid due to the negligible accumulation of energy in the gas and

the high heat transfer coefficient between the gas and solid. This figure along
1. The profiles shown in Figure 3.4·8 are at the radial collocation point r c·
 -232-

with further evidence presented later in this thesis shows that the dynamic

behavior of the reactor is dominated by the catalyst and thermal well thermal

behavior.

Many other disturbances have been simulated, leading to an Wlderstanding

of the control complexities of the system. Many of these simulations are shown

in Sections 3.6 and 3.7 and are therefore not reproduced here. Axial tempera-

ture measurements have been found to be essential in determining and control-

ling the behavior of the process. Outlet temperatures alone cannot provide ade-

quate information for control of the reactor.

3.3.3 Effects of Reactor Operating Conditions

One of the major purposes of accurate mathematical modeling of the reac-

tor bed is to study the effects of various operating conditions on the behavior of

the reactor, thus allowing process optimization and insight into the perfor-

mance of the system under changes in various input parameters. This enables

careful design of control structures for the experimental system without

significant a priori experimentation.

Figure 3.4-10 shows the effect of the inlet gas temperature on both the

outlet gas temperature and CO conversion for the reversible kinetics given by

Equations (3.3-:) and (3.3-3), and Figure 3.4-11 shows the effects on the steady

state axial gas temperature profiles. As expected. the conversion and the 'hot

spot' temperature increase with increasing inlet gas temperature. However

under many conditions, the outlet gas temperature is inversely related to inlet

gas temperature. An increase in the inlet gas temperature produces a decrease

in the outlet gas temperature2 as a result of the shifting of the 'hot spot' down

2. Tne so-called 'wrong-way' behavior.

 ~ 233

the bed (see Figure 3.4-7 also). However as shown in Figure 3.4-10, this is not

t:rue throughout the possible operating regimes. Obviously, this behavior can

lead to significant control difficulties it the control design is based on the outlet

gas temperature as is often the case.

The steady state axial concentration profiles for an inlet gas temperature

of 623° K are shown in Figure 3.4-12. As before, the steam-shift reaction leads to

a slight formation of C02 in the early part of the reactor due to the presence of

a large amount of CO. However as the CO is rapidly depleted in methane forma-

tion, the steam-shift reaction reverses. Due to the higher temperatures Within

the reactor, the rates for both reactions are much higher than for the standard

Type I conditions.

Figure 3.4-10 also shows the tremendous etiect that the inlet velocity or

flowrate has on conversion and temperatures. Figure 3.4-13 shows the steady

state axial gas temperature profiles at various inlet gas velocities. A sudden

drop in the inlet fiowrate would cause the 'hot spot' to become much more pro-

nounced and to shift towards the entrance of the bed. The lower fiowrates also

lead to higher conversions (not shown).

Finally, the effects of the inlet CO concentration and cooling fluid tempera-

ture on the CO conversion and outlet gas temperature are shown in Figures 3.4-

14 and 3.4-15. An increase in either the inlet CO concentration or the Dowtherm

temperature 3 increases the conversion and the outlet gas temperature. Not too

surprisingly, the relationship between the outlet gas temperature and cooling

fluid temperature is nearly linear since for these conditions the reaction is

nearly complete in the first half of the reactor and the second half acts as a

heat exchanger to cool the gas.

3. This would be the result of increasing the nitrogen pres~mre in the Dowtherm condenser.
 -234-

3.3.4 Radial Concentration Analysis

Radial gradients are generally ignored in dynamic analyses of packed bed

reactors, mainly due to the increased complexity of the resulting model. As

pointed out by Jutan et al. (1977), in the only other major dynamic packed bed

reactor study that incorporates radial profiles, radial gradients are important in

industrial processes where wall cooling is required for safety or control. The ori-

ginal mathematical model developed in the proceeding section includes com-

plete radial analysis of both the temperature and concentration profiles. How-

ever, explicit solution of the resulting partial differential equation model is not

feasible, and discretization is needed to reduce the model to a form suitable for

numerical solution. This reduction used the method of orthogonal collocation

in both the radial and axial dimensions.

The radial discretization must then account for the projected radial

profiles. In the analysis performed in the lasl seclion. one interior radial collo-

cation point was used. This along with the two radial boundary values at the

thermal well and cooling wall resulted in an inherent assumption of quadratic

radial gradients. An examination of experimental measurements (Yalstar et al..

1975) and typical radial profiles (Finlayson, 1971) for similar reactors indicates

that radial temperature profiles can adequately be represented by a quadratic

function of radial position. This quadratic representation is preferable to one of

higher order since the dimensionality of the system is minimized through the

use of only one interior collocation point.

However, a quadratic representation of the radial concentration profile may

not be adequate since application or zero .tlux conditions at the inner thermal

well and outer cooling wall with a quadratic profile reduces to an assumption of

constant radial concentrations. Although additional radial collocation points

 -235-

greatly increase the dimensionality of the resulting model, they may be neces-

sary to accurately express the radial concentration profiles. This section con-

siders the problem in detail by comparing simulations with additional radial col-

location.

The original model is discretized in the radial dimension using orthogonal

collocation with multiple interior collocation points for the concentration

profiles and a single radial collocation point for the gas and catalyst tempera-

tures. For example, with two interior collocation points, the assumed radial

concentration profile is

(3.4-1)

This profile must satisfy the boundary conditions

(3.4-2)

The profile must also be exact at the collocation points, r =t;Q 0 , rc 1 , rc 2 , and 1.0:

Yi 0 = do + d1 'Po + d2cp~ + dstf'&

Ytc = do + dtrc 1 + ~r;1 + dar~1
1
(3.4-3)
Yica =do+ d1rc 2 + d2r;2 + dsrg2
Y1. 1 = do + dt + d2 + ds

After rearranging and eliminating Yio and Yi 1•

0 1 2 3 do 0
0 1 2t;Qo 3t;Q§ dl 0
2 s = (3.4-4)
1 rcl rcl rcl d2 Yir
cl
2
1 rea rc2 r3 da
Ce Y;,r
02

The expressions for di(~() can be obtained as:

 -236-

(3.4-5)

(3.4-6)

(3.4-7)

(3.4-8)

where a =[(rk', -rl,l - ~ (r~.-r~,) (1 +l"ol + 31"o(r,,-r,,) r(3. 4-9)

These radial collocation solutions are then substituted into the partial

differential equations, using the relationships

Byt 2
&t =d 1 ~ + 2d~r + 3d3,r
a2y·
.--!....: 2~ + 6d r
1Jr2 ' s,
Bp, = _ M,Pt B@,
(3.4-10)
ar a: ar
Bc5 -o 8yl - By2
ttr = -xco iJr - Xcog 7ir'
CJ2o _ -oB2 Y1 - fPy2
c;r2 - -xco c;r2 -- Xco~ ar2

ae
where tJrg is obtained from the one point radial collocation of the gas tempera-

tures. The concentration equations are then valid at both radial collocation

points rc 1 and rez since the residuals will be zero at these points. The full radial
 -237-

concentration profile can be reconstructed from Equations (3.4-1) and (3.4-5)

through (3.4-9). The energy balances for the gas, catalyst, and thermal well and

the continuity equation are unchanged.

Axial collocation and numerical solution is performed in an identical

manner as before with this larger system of equations. The size of the

differential equation model has been increased by 2N nonlinear ordinary

differential equations, where N is the number of axial collocation points. Note

that for each additional radial collocation point for the concentrations, the

model will be increased by 2N nonlinear ordinary differential equations.

Simulations were then performed with this model and compared to those

using the earlier model. In the simulations, the bulk concentration can then be

obtained by integrating the radial profiles:

Bulk Concentration (3.4-11)

This allows direct comparison with previous simulations. Figure 3.4-16 shows a

comparison of the axial gas temperature and bulk concentration profiles for the

reactor with standard Type I operating conditions using the original model with

no radial concentration gradient (i.e., infinite radial diffusion) and using the new

model with a representative radial mass Peclet number of 2.0 and with a radial

mass Peclet number of oo (i.e., no radial diffusion). Figure 3.4-17 shows a similar

comparison with Type II operating conditions.

Figure 3.4-18 shows the radial temperature profiles at the reactor outlet.

The radial and axial temperature and bulk concentration profiles are effectively

not influenced by these modeling differences. Figure 3.4-19 shows the radial con-

centration profiles at ( = 0.38 and at the reactor outlet. Even with very high

Peclet numbers, the differences between the radial concentration profile across
 -238-

the relatively small bed and the assumed uniform profile are minimal. Definitely

under planned operating conditions with small Peclet numbers, there is no

benefit to increasing the number of radial collocation poinls, especially in light

of the increased dimensionality of the resulting system.

3.3.5 Adiabatic Analysis

Modeling of the packed bed catalytic reactor under adiabatic operation sim-

ply involves a slight modification of the boundary conditions for the catalyst and

gas energy balances. A zero flux condition should be used at the outer reactor

wall. This can be accomplished us1ng the programs developed in this work sim-

ply by setting the outer wall heat transfer coefficients. hws and hwg (or

corresponding Biot numbers), equal to zero. Simulations under adiabatic opera-

tion do not significantly alter any of the conclusions presented throughout thls

work and are often used for verification of worst-case nonadiabatic operation.

Figure 3.4-20 shows the adiabatic steady state temperature and concentration

profiles for standard Type 1 operating conditions. As expected, the temperature

rise through the bed is more dramatic than in the non-adiabatic case, leading to

much higher conversions through the bed. Due to the lack of any cooling, no

'hot spot' develops and all heat generated by the reaction is removed from the

bed by the product gas, resulting in a very high temperature rise (200° K) within

the bed. Such high temperatures are undesirable since they can significantly

increase catalyst deactivation. Also note that in the adiabatic analysis the tem-

perature difference between the gas and catalyst is negligible. Although the heat

transfer coefficient between the gas and solid are as large as in the nonadiabatic

analysis, major differences exist during nonadiabatic operation between the

radial heat transfer through each phase. 4 The results shown in 3.4-20a along
4. The heat tran.s:!'er coefficient from the solid to the outer wall and the radial conduction in the
 -239-

with other simulations indicates that a homogeneous analysis of the bed may be

adequate for adiabatic analyses.

Figure 3.4-20b shows that CO conversion is much higher under adiabatic

operation due to the higher bad temperatures. Note that the conversion of the

C02 becomes important as soon as CO is nearly depleted. The 'rippling' in the

C02 curve is a result of the axial orthogonal collocation. 15 Numerical solution

problems such as this will be discussed in Section 3.5.

3.3.6 Importance of Thermal Well

The mathematical model developed in this work allows for an analysis of

the effects of a central axial thermal well. Although the presence of the well was

found to have little effect on the concentration profiles, it significantly alters the

transient temperature response of the reactor bed, since its large thermal capa-

cllance increases the thermal time constant of the bed. Figure 3.4-21 shows the

radial gas and solid temperature profiles at the reactor exit (z = L) during reac-

tor start-up. As shown, the steady state profiles are similar, although conduc-

tion along the thermal well slightly alters the surrounding gas temperature. The

transient behavior exemplifies the slow response of the thermal well. It can be

seen that the response of the exit temperature profile is rapid without the well,

With steady state being approached in under one minute; whereas, the presence

of the well introduces a finite heat sink into the reactor center that slowly

absorbs some of the heat produced. Over ten minutes are necessary to

approach steady state with an eighth-inch diameter well. This is further seen in

Figure 3.4-22 where the temporal behavior of the catalyst and gas temperature

solid is greater than the correspondi.ng parameters in the gas phase.

5. Although only 6 axial collocaLlou poinl.8 a.re used in most simulatioil3, 12 points were necessa.ry
here and even then the results are less than optimum..
 -240

at the reactor exit are compared With and without the presence of the thermal

well. These figures also show that the temperatures in the reactor center are

higher without the thermal well as would be expected. However, Figure 3.4-23

shows that the well has very little effect on the axial concentration profiles, and

overall conversion is only slightly affected.

Although the thermal well increases the thermal capacitance of the reactor

bed and reduces the reaction volume, these effects alone cannot account for the

dramatic increase in the thermal time constants of the bed with a thermal well. 6

Whereas the dynamics of the catalyst particle temperatures are very fast due to

the heat generation on the particles from the exothermic reactions. the dynam-

ics of the thermal well temperatures are much slower since the heat is gen-

erated on the surrounding particles. Much of this energy is transferred to the

outer cooling jacket or out of the bed with the gas phase Only a small portion is

transferred to the thermal well and even this through relatively slow mechan-

isms, thus resulting in the slow dynamic behavior of the well.

Since a thermal well is sometimes used m mdustry to obtain temperature

measurements for process monitoring or control, we note the importance of

incorporating even a relatively small well into the modeling and control analysis.

Basing decisions on measured temperatures Wlthin an ax1al thermal well and a

model without a thermal well (i.e .. assuming that the measured temperatures

are the predicted temperatures at the center of the bed) can be dangerous.

Simple modifications to the model development presented in this work

allow for simulating systems that do not include central thermal wells. 1n such

cases, the dimensionality of the mathematical system is reduced by N, since one

partial differential equation is eliminated. The modeling programs 'developed in

6. Due to the relatively small radius of the well, the thermal capacitance (Vpeo) o~ the bed with the
thermal well (16 cal/ 0 K) is only slightly higher than that without (14 caft"K) a.."ld the loss in
reaction volume is only about 2~.
 -241-

this work can be used directly for systems without thermal wells simply by set-

ting the parameters for the well diameter and the thermal well heat transfer

coefficients, hts and htg (or corresponding Biot numbers), equal to zero.

3.3.7 D:ispennon Effecl:s

Much work has been focused on the significance of dispersion terms in the

transient material and energy equations for packed bed reactors. In general.

axial diffusion of mass and energy and radial diffusion of mass have been

neglected in comparison with convective terms in most packed bed reactor stu-

dies (Carberry and Wendel, 1963; De Wasch and Froment, 1971; Hlavacek, 1970;

Heiberg et al., 1971: Jutan et al. 1977: Valstar et al., 1975)

For packed bed reactors, Carberry and Wendel (1963), Hlavacek and Marek

( 1966), and Carberry and Butt ( 1975) report that axial dispersion effects are

negligible if the reactor length is sufficient. These and other researchers (Mears.

1976; Young and Finlayson, 1973) have developed criteria based on the reactor

length for conditions where the axial dispersion can safely be neglected. Since

the tube length/pellet diameter for the experimental methanation reactor is

about 150, just at the limit of most of the published criteria for neglecting axial

heat dispersion, careful analysis of these dispersion effects is needed. However,

neglecting axial mass dispersion should be a safe assumption since the tube

length/pellet diameter for the experimental system is well above the value of 50

generally recommended. Radial mass dispersion can also be neglected and is

inherently eliminated from our model since the one point radial collocation

results in a constant radial concentration profile. However, the radial d.i.fi'usion

of energy must be retained since it is one of the most important physical

processes influencing the dynamic and steady state behavior of the system by
 -242-

governing the radial fl.ow or energy through the bed to the cooling wall.

Simulations using the full model were then used to study the necessity of

these dispersion effects and to verify some of the common assumptions. A wide

range of Peclet numbers were used in the simulations. The actual values for the

methanation reactor will have to be estimated from preliminary heat transfer

experiments in the reactor (Section 4.2). Based on extensive previous studies.

the radial gas Peclet number should range from 5 to 10 and the axial gas Peclet

numbers from 0.5 to 2.0 (Carberry, 1976). For completeness, our simulations

used Peclet numbers over a much wider range.

Comparison of steady state profiles for Type I conditions (Figure 3.4-24)

shows that neglecting axial ma~~ difl'u~ion has very little effect on the tempera-

ture and concentration profiles even though the axial gradients are significant.

However, Figure 3.4-25 shows that neglecting the axial thermal dispers10n in the

gas docs affect the solution profiles. The axial temperature profiles are offset

and the concentration profiles are shifted slightly. Further simulations show

minimal effect of neglecting the axial conduction of energy in the solid.

Figure 3.4-26 shows that, for standard Type II operating conditions, neglect-

ing the axial thermal dispersion leads to instabilities in the orthogonal colloca-

tion solution with six collocation points. Although the solutions at the colloca-

tion points are similar, neglecting axial dispersion of heat leads to some 'rip-

pling' in the axial temperature profiles. However, the concentration profiles are

nearly unaffected. This 'rippling' is a result of the axial orthogonal collocation

and can be reduced by increasing the number of collocation points. However,

this leads to a substantial increase in model dimensionality and thus solution

time. Obviously, the axial thermal diffusion has a stabilizing effect o'n the numer-

ical solution using orthogonal collocation and actually damps the behavior of
 - 2~3-

the solution polynomial.

Finally, Figure 3.4-27 shows the effect of neglecting axial diffusion on the

dynamic simulations of the reactor under start-up operation with Type II condi-

tions. Significant differences are apparent.

The reduction in solution time for these simulations is minimal nnd in some

cases the elimination of the axial diffusion terms actually increases the solution

time. Simulations show that neglecting the axial dispersion of mass has little

effect on numerical computation time; whereas, eliminating axial dispersion or

energy may significantly increase computation time and only rarely decreases it

substantially.

Thus our analysis of the effects of dispersion on the simulated behav10r of

Lhe rea.ctor along with the work by Bonvin (1980) shows the necessity of mclud-

ing the thermal diffusion terms. Simulations here verify that the numerical

stability of the model is greatly enhanced by retaming these dispersive effects

and that, although minor additional effort may be necessary in the model

development, the numerical solution time may actually be reduced by retaining

some of these terms. Furthermore, simulations verify that the axial dispersion

of mass can usually be neglected. although the inclusion a! these terms in the

model introduces little complication in the collocation solution; whereas, other

solution procedures are often significantly hindered by these second derivative

terms. The radial dispersion of mass seemingly has little effect on either the

simulated results or solution times and can safely be neglected. This conclus10n

is based on the radial concentration analysis presented in Section 3.4.5, since

the standard mathematical solution that uses one point radial collocation in

thts work effectively assumes no radial concentration gradient.

 -244-

970
-510

'"o -a 950
X ··~- X

"~
z z
8 -520 0
,_ t-
930
u u
a: a:
w w
IT a:

~
~
lJ.._ 910'"'
0 -530 r- lJ.._
0
f-
a:
I
I
f-
u.
w w
I I 890 1-

-540
870

-550 '---.1....--~-~-J__-.J.......____.j 850'----~--~--~--~-------

500 600 700 800 500 6CO 700 800
TEMPERATURE WEG Kl TEMPERATURE (QEG Kl

(a) (b)

Figure 3. 4~ 1
Temperature Dependence of Heats of Reaction
(a. Methanation, b. Steam-Shift)
(Source: United Catalysts Inc.)
 - 2-±5-

2lJ. 0

20

I·;
-1
I- I-
z z
a: a:
I- I-
(j') 16 (j')
z z
0 0
u u -2
:,;:: I :,;::
::::J
I ::::J
12 I
a: a:
ti:l
I
,1 Cil
_J -3
....J

.:.J I
/ ::I
e
0
w I 0
I.U

z
_J

lJ.
I
,I :z
.....J
-lJ.

QL-----~----~----~--~ -5~----~----~----~--~
0.12 0. 16 z 0.20 0.12 16 o. 2 0.20
1/T (QEG Kl M 10 1/T COEG Kl JoElO

(a) (b)

Figure 3.4-2
Temperature Dependence of Equilibrium Constants
(a. Methanation, b. Steam-Shift)
(Source: United Catalysts Inc.)
 -246-

0.1.1 3. 10

"'' o
'o
%
~ 0.3 3.05
(\j
0 D
u u
C02
z z:
0 D
1-
0.2 3.00
1--
u u
cr: a:
a: co cc
LL
LL
LLJ 2.95
.....I
0. 1 LLJ
__j
0 0
:L :L

0.0 2.90
0.0 0.2 0.1.1 0.6 0.8 1.0
AXIAL POSITION, Z/L

(a)
675

650
~

t:J
LLJ
0
625
LLJ
cc
::::>
1-
cr: 600
cc
LLJ
n
:L
w 575
1--

sso~~--~--~--~--~--~--~~~~--~
o.o 0.2 O.Y 0.5 0.8 1.0
AXIAL POSITION. l/L

(b)

Figure 3.4-3
Steady State Axial Profiles
Standard Type I Operating Conditions

a) Concentration Profiles Based on Inlet Total Moles

b)· Temperature Profiles at r = Ro. r c• R1
- Solid Phase - - - - Gas Phase
 -247-

675
z = 0.38L

::s:::
650 ----
(._J
w
0
625
z= L
w
a:
:;::)
----
I-
cr: 600
a:
w
o_
L
w 575 z =0
I-

550L---~--~--~----~--~--~--~--~--~--~
0.0 0.2 0.4 0.6 0.8 1.0
RROIRL POSITION. R/Rl

Figure 3. 4-4
Steady State Radial Temperature Profiles
Standard Type 1 Operating Conditions
- - Solid Phase - - - - Gas Phase
 -248-

0.8 !. 60

0. 7
"''o
'o ,......
)1:
X 1. 55
C\J
0 0.6 0
u u
z z
0
0
I-
0.5 I. 50
I-
u w
a: a:
a: a:
lJ... O.ll Lt..
lJ.J I. CIS
...J LL.I
_j
0
::£: 0.3 0
::£:

0.2 I. 40
o.o 0.2 0.4 0.6 0.6 1.0
AXIAL POSITION. ZIL

(a.)

650

~
625
r...::l
lJ.J
0

w
a: 600
:::J
f-
rr
a:
lJ.J
CL
:E 575
LiJ
1-

0.2 0.4 0.6 0.6 1.0

AXIAL POSITION. Z/L

(b)

Figure 3.4-5
Steady State Axial Protiles
Standard Type II Operating Conditions

a) Concentration Profiles Based on Jnlet Total Moles

b) Temperature Profiles at r =
Ro. rc, R1
- Solid Phase - - • - Gas Phase
 -249-

600
z=o~­
=
z 0. 17L

w
----
a:
::J
1-
a: z = 0
a: 575
w
CL
::E
w
1-

550~--~--~---L---L--~--~~--L---~--~--~
0.0 0.2 0.4 0.6 0.8 1.0
RRDIRL P~SITI~N. R/Rl

Figure 3.4-6
Steady State Radial Temperature Profiles
Standard Type II Operating Conditions
- - Solid Phase - - - - Gas Phase
 -250-

675

::X:: 650
l:)
w 625
0

w
a:: 600
::J
I-
cr:
a:: 575
w
(L
:E
w 550
I-

c.n
cr: 525
l:)

500
o.o 0.2 O.l.l 0.6 0.8 1.0
AXIAL PClSITIClN, 2/L

Figure 3.4-7
Disturbance (10% drop in Inlet Gas Temperature) Analysis
Transient Axial Temperature Profiles
Standard Type I Operating Conditions
 -251-

650
~

c._:)
w
0

w
a:
:::::>
t- 600
a:
a:
w
0...
::E
w
t-

550L---~--L---~--~--~--~--~---L---L--~
0.0 0.2 O.l! .0.6 0.8 1.0
AXIAL POSITION, Z/L

Figure 3.4-8
Axial Gas and Catalyst Temperatures during Start-up
Standard Type I Operating Conditions
- - Solid Phase - - - - Gas Phase
 -252-

z =l - -- - - - -

w
- ______
- - - - - - - -_-
/
a: 625 ......
::::> /
_.:::,___
1-
a:
,....,.--....
a: '// ,; ,. z = 0.38L
w /
/'
Q_
L
w
1-

600~~~--~-L~--~-L--L-~~--L--L--~J-_J

0 20 l!O 60 80 100 120

TIME (SECl

Figure 3. 4~9
Temperatures A!ter an Input Step Change
Standard Type II Operating Conditions
(Tw from 573° K to 593° K)
Solid Phase - - - Gas Phase ---- Thermal Well
 -253-

650 100

,.,...-...--
645 ./"""" / 90
ug = /
/ /
/
640 / /u
/ /
9
= 100 80
::t:: / /
L:)
w
635 / /
a / /
70

w 630
/ /
,...,
;;-..:
/
a: / 60
:::::> z
1-
a: 625 / 0
........
a: / {f)
w 50 a:
a_
:E
620
/ w
>
w / z
1- 0
(f')
40 u
a: 615 u = 75
L:) g 0
u
1- 30
LLl
.....) 610
1-
:::::>
0 20
605

600 10

595 0
400 450 SOD 550 600 650
INLET GRS TEMPERATURE lDEG Kl

Figura 3.4-10
Effects of Inlet Gas Temperature
Standard Type I Operating Conditions
- Outlet Gas Temperature
- - - CO Conversion
 750

:X::
(.!)
700
w
Cl

w 650
0::
=:l
1-
a:
a: 600
w
a...
:E
w
1-
(f)
550
a:
(.!)

500
0.0 0.2 0.1! 0.6 0.8 1.0
AXIAL PCl5ITIClN, Z/L

Figure 3.4-11
Effects of Inlet Gas Temperature
Steady State Axial Temperature Profiles
Standard Type I Operating Conditions
 -255-

0.4 3.05

-' o 3.03
N
'o
......
X
X 0.3
C\J
0
u 0
u
z 3.01
0
z:
....... 0
0.2
1-
1-
u u
cr: 2.99 cr:
0:: 0::
l..!....
LL..
w co
.....J
0. 1 w
2.97 .....J
0
0
L: L:

0.0 2.95
0.0 0.2 o.tt 0.6 O.B 1.0
AXIAL POSITION. Z/L

Figure 3.4-12
Axial Steady State Concentration Profiles for Tc = 623° K
Standard Type I Operating Conditions
 -256-

700

~
675
<..:::)
w
a
650
w
a:
:::>
1- 625
a:
a:
w
Q_
L 600
w
1-
(f)
a: 575
(.!)

550
0.0 0.2 0.1! 0.6 0.8 1.0
AXIAL POSITION. Z/l

Figure 3.4-13
Effects of Inlet Gas Velocity
Steady State Axial Gas Temperature Profiles
Standard Type I Operating Conditions
 -257-

675 100

/
/ ------ 95
:::s:::
/
650
c._:)
/
w
Cl
I 90

I ;-...::
w
a: I 85
::;:::)
r-
a:
625
I z
D
a:
w I
I .......
(f)

CL
8U a:
~
w
w >
r- z
600 0
(f) 75 u
a: D
c._:)
u
r-
w 70
......J
r- 575
::::?
D
65

sso~--~------~----~--~----~----~--~----~60
o.oo 0.02 0.0~ 0.06 0.06
INLET CD MOLE FRACTION

Figure 3.4-14
Effects of Inlet CO Concentration
Standard Type I Operating Conditions
- Outlet Gas Temperature
- - - CO Conversion
 -258

675 100

:X::

c..::>
650
/
/
/
/
-- --- 95

w 90
c::l /
/ ~
w /
a: 85
::::)
I-
625 / :z
cc / D
a: .......
(./)
w 80 0::
(L
::L w
w >
1- :z
600 D
(./) 75 u
cc D
c..::>
u
I-
w 70
_j
I-
::::)
575
D
65

550 60
500 550 600 650
CDDLING FLUID TEMPERATURE WEG K)

Figure 3.4-15
Effects or Cooling Fluid Temperature
Standard Type I Operating Conditions
- Outlet Gas Temperature
- - - CO Conversion
 -259-

L::l
w 625
0

w
a:
:::l
1-- 600
a::
a:
w
a...
:r
w
I- 575

550~--L-~~~--~--~--~--~--~--~--~
0.0 0.2 0.4 0.6 0.8 1.0
AXIAL POSITION, Z/L

(a)

...
'o 'o
3.05
D
w
z co
D
z
D
1--
1--
w w
a: a:
a: a:
l.1..
l.l...

0. 1 2.95 w
_j
D
:I:

O.OL-~~~--~--~--~--~--~--~--~--~2.90
0.0 0.2 0.4 0.6 0.8 1.0
AXIAL POSITION, Z/L

(b)

Figure 3.4-16
Multipoint Radial Concentration Collocation
Axial Gas Temperatures and Bulk Concentration Profiles
Standard Type I Operating Conditions

Original Model - Infinite Radial Diffusion

New Model- Radial Mass Peclet Number = 2.0
- - - - New Model - Radial Mass Peclct Number = ""'
 -260-

650

::.::::
L'J
w 625
0

w
a:
::::J
1- GOO
a:
a:
w
CL
:I:
w
1- 575
(/)
a:
L'J

550
0.0 0.2 O.ll 0.6 0.8 1.0
RXlRL POSITIClN, l/L

(a)

0. 10

N
'o
0.08 X
0 1. 55
w N
C0 2 0
z =--~ w
0 0.06
1-
z
0
w J. 50
a: 1-
a: (_)
iJ_ O.Oll a:
w a:
iJ_
.....J
0 !.liS w
:t:: .....J
0.02 0
:t::

0.00 l.l!O
o.o 0.2 O.ll 0.6 0.8 1.0
AXIAL PeJS IT I CJN, Z/L

(b)

Figure 3.4-17
Multipoint Radial Concentration Collocation
Axial Gas Temperatures and Bulk Concentration Profiles
Standard Type II Operating Conditions

- Original Model - Infinite Radial Diffusion

- - - New Model - Radial Mass Peclet Number = 2.0
---- New Model- Radial Mass Peclet Number = oo
 -261-

620

610

w
a: 600
:::::>
f-
a:
a: 590
l.i..J
0...
:L:
w
t- 580

570~--~--~--~--~--~--~~--L---~--~--~
0.0 0.2 O,ij 0.6 0.8 1.0
RADIAL POSITION, R/Rl

Figure 3.4-18
Multipoint Radial Concentration Collocation
Radial Gas and Catalyst Temperature Profiles at Reactor Outlet
Standard Type II Operating Conditions

- Original Model - Infinite Radial Diffusion

- - - New Model - Radial Mass Peclet Number = 2.0
--- - New Model - Radial Mass Peclet Number :::: oc
 -262

0.60 T l

-' a...... 0.55

:.:
D
z = 0.38L
u 0.50 --=---=-
z
-
D
1-
u
0.45
a: -
a:
LJ.... 0.40
w z = L
....J
D
-
-----=- .....----=-- ------ -
:::r:: 0.35
- --
0.30 I

0.0 0.2 0.14 0.6 O.B 1.0

RADIAL PCJSITICJN, R/Rl

Figure 3.4-19
Multipoinl Radial Concentration Collocation
Radial Concentration Profiles
Standard Type I Operating Conditions

---....... Original Model- Infinite Radial Diffusion

- - - New Model - Radial Mass Peclet Number = 2.0
- - - - New Model - Radial Mass Peclet Number = co
 -263-

750
::.:::
t.:)
w
Cl
700
w
a:
:::>
t-
a:: 650
a:
w
0....
:r:
I.I.J 600
t-

550
0.0 0.2 0.1! 0.6 0.6 1.0
AXIAL PDSITIClN, Z/L

(a)

O.ij

N
"io 'o
J: 0.3 :c:
3.05
(\J
0
u 0
u

--
z C0 2
0 z
0
0.2 3.00
t·
u
""'- ............ 1-
u
a:: '\
a: co
----~ a::
a:
1.!...
w
"""" ""'
1.!...

~
_J o. 1 2.95 w
_J
0
:r: 0
"'-..

-- --
X:

o. 0.
' ""'
- 2.90
0.0 0.2 O.ij 0.6 0.6 1.0
AXIAL POSITION, ZIL

(b)

Figure 3.4-20
Adiabatic Steady State Axial Profiles
Standard Type I Operating Conditions
a) Temperatures ( - Catalyst - ..... - Gas)
b) CO and C02 Concentrations(~~ Nonadiabatic - - - Adiabatic)
 -264-

630
Lso, 100, steady state
X:
620
L:l
w
- -
Cl 610

-r------
a: 600
~
1--
25
a::
a: 590 -~
w
ll...
:L
w
I-
570

560~--~--~--~--~--~--~--~--~--~~
0.0 0.2 0.4 0.6 0.8 1.0
RADIAL POSITION, R/Rl

(a)

530 _ _[_so. 100, steady state

X::
620 --- steady state
L:l
w
Cl 610

w
cc 600
:J
1-- 25
a::
a: 590 ~-;--.. 10 s;;---- _
w --__;;::.-=---..._.,.'='---
~.. 10 sec
ll...
::E 580
w
I-
570

560~--~--~--~--~--~--~--~--~--._~
0.0 0.2 0.4 0.6 0.8 1.0
RADIAL PDSITIDN, R/Rl

(b)

Figure 3.4-21
Effect of Thermal Well on Transient Radial Temperature Profiles
Standard Type II Operating Conditions
- - . 1/B"Thermal Well - - - No Thermal Well

a) Gas Temperatures
b) Solid Temperatures
 -265-

625
--------
/"" --------
~
'l
I.J...I
a: 600 ~
=:>
f-
a:
a:
w
a_
:E 575
I.J...I
t-

550~--~--~--~--~--~--~--~--~--~--~
0 20 40 50 80 100
TIME (SECl

Figure 3 4-22
Dynamic Behavior of Exit Temperatures
Standard Type II Operating Conditions
- - 1/8" Thermal Well - - - No Thermal Well
 -266-

0.8 1. 60

... N
'o
'o 0. 7
....... X
X 1. 55
(\J
D D
u 0.6
u
z z
D Q
....... 0.5 1. 50
1-- I-
u u
cr: cr:
a: a:
LL 0.4 LL
w 1.1.!5 w
_J _J
D

-··~1."0
0.3 D
::L: L

_____ j __
0.2
0.0 0.2 O.ll 0.6 0.8 1.0
AXIAL PCJSITICJN, Z/L

Figure 3. 4-23
Steady State Axial Concentration Profiles
Standard Type II Operating Conditions
- - 1/B"Thermal Well - - - No Thermal Well
 -267-

w
a:
~
t-
o:
a:
w
Q_
~
w
t-
U1
a:
~

550
0.0 0.2 0.4 0.6 0.8 1.0
RXIRL PClSITIClN, Z/L

(a)

0.4 3. 10

'o "''o
X
X 0.3 3.05
(\J
D
w D
w
:z C0 2
D :z
D
t-
0.2 3.00
w t-
a: w
a: co a:
LL a:
LL
w 0. 1 2.95 w
...J
D ...J
~
D
~

o.o~--~~--~--~--~--~--~---L--~--~2.90
0.0 0.2 0.4 0.6 0.6 1.0
AXIAL PCJSITIDN, Z/L

(b)

Figure 3.4~24
Steady State Axial Profiles - Neglecting Axial Mass Diffusion
Standard Type I Operating Conditions
Original Model No Axial Mass Diffusion

a) Gas Temperature
b) CO and C02 Concentrations
 -268-

l.:)
w
a
w
a:
::J
t--
a::
a:
w
a...
::t
w
t--

lf)
a:
l.:)

550
0.0 0.2 0.4 0.6 0.8 1.0
AXIAL POSITION, Z/L

(a)

0.16 3. 10

"'' o
'a
X
X 0.3 3.05
(\J
0 0
w w
z z
0 Q
0.2 3.00
t-- t--
w w
a: a:
a: a:
LL LL
w 0. l 2.95 I.J.J
_..J _..J
0 0
:i:: z:

0.0 2.90
o.o 0.2 O.ll 0.6 a. a 1.0
AXIAL POSITION, Z/L

(b)

Figure 3.4-25
Steady Stale Axial Profiles -Neglecting Axial Heat Diffusion
Standard Type I Operating Conditions
Original Model No Axial Heat Di1fusion

a) Gas Temperature
b) CO and C02 Concentrations
 -269-

600
I.J..J
a:
:::>
t-
a:
cr:
I.J..J
CL
::L 575
UJ
t-

U'"l
a:
~

550
0.0 0.2 O.ll 0.6 0.8 1.0
RXIRL P~SITI~N. Z/L

(a)

0.8 I. 60

"''o
'o...... 0. 7
X
X 1. 55
("\J
C)
u 0.6 D
u
:z z:
C)
C)

t-
0.5 1. so
t-
u u
a: a:
cr: cr:
lJ._ O.ll lJ._

I.J..J
....J
l.ll5 I.J..J
D ....J
C)
::L 0.3 ::L

0.2 1. LID
0.0 0.2 O.ll 0.6 0.8 1.0
AXIAL POSITION, Z/L

(b)
Figure 3.4-26
Steady State Axial Profiles - Neglecting Axial Heat Diffusion
Standard Type II Operating Conditions
Original Model No Axial Heat Ditl'usion

a) Gas Temperature
b) CO and C02 Concentrations
 -270-

625

:l::::
<..:)
w
0
600
w
cr.::
::;::)
1- t = 10 sec
a:
a:
u..l
a_
~
575
I..L...I
1-
(f)
cr:
<..:)

550
0.0 0.2 0.1! 0.6 0.8 1.0
AXIAL PCJSITICJN. Z/L

Figure 3.4-Z7
Dynamic Start-up Simulations - Neglecting Axial Heat Diffusion
Standard Type II Operating Conditions
- - Original Model - - - No Axial Heat Diffusion
 -271-

3.5 OCFE MODEL ANALYSIS

The reaction rate expression for the methanation reaction, Equation (3.3-

1), used throughout much of this work is based on empirical work by Lee (1973)

and Yatcha ( 1976). This expression, along with the rate equation. Equation (3.3-

3), for the steam-shift reaction given by Moe (1962) provides valuable insight

into the modeling of multiple reaction systems. Both reactions are reversible

and are described by complex nonlinear rate expressions. Due to the reversibil-

ity of the reactions and the equilibrium constraints included in the rate expres-

sions, severe temperature and concentration profiles are generally not observed.

The orthogonal collocation procedure used for the numerical solution of the

resulting mathematical model shows very little numerical instability and allows

for rapid solution of even the most difficult profiles simply by increasing the

number of collocation points.

However, in many chemical systems including packed bed reactors with

rapid nonreversible kinetics, the interesting features of the solution are

confined to a very small region where the solution profile changes rapidly. The

orthogonal collocation method may become unwieldy because a large number of

collocation points may be needed so that enough are plac.ed within this region to

provide an accurate representation of the solution.

A significant drawback of the orthogonal collocation technique is then its

inability to accurately define profiles with very sharp gradients or abrupt

changes, since the technique requires fitting a single polynomial to the entire

profile. Under such cases, although the solution will be exact at the collocation

points, significant oscillations in the profile are observed, and the obvious choice

would be to resort to a finite difference procedure which considers each interval

separately using first and second derivative matching conditions on the bound-
 272-

aries or to a finite element approach. Accurate representation of the solution is

then possible even under extremely steep gradients simply by increasing the

number of grid points. However, this technique can easily become numerically

prohibitive due to the extremely high dimensionality of the resulting mathemat-

ical system. A compromise between the orthogonal collocation (OC) and finite

dl.1.':ference techniques has been proposed by Carey and Finlayson (1975). This

procedure is called Orthogonal Collocation on Finite Elements (OCFE).

3.5.1 Formulation of OCFE Technique

Although most or the preliminary analyses in this modeling work used the

rate expressions by Lee ( 1973), Moe ( 1962), and Vatcha ( 1976). all significant

conclusions were verified using the rate expression of Strand ( 1984) and ignor-

ing the steam-shlft reaction. An important difference between the two kinetic

descriptions is the sharpness of the resulting profiles. The reaction kinetics

determined by Strand (1984) are much faster than that proposed by Vatcha

( 1976) and can lead to rapid complete conversion Within the reactor bed. For

such kinetics, the numerical solution procedure described in Section 3.3.5 can

have difficulties during solution of the steady state profiles.

Specifically, numerical solution of the expected steady state profiles with

the new kinetics often requires excellent initial estimates, and in many cases

even then has numerical convergence difficulties. This is often due to the rapid

complete reaction of the carbon monoxide early in the reactor bed leading to

steep temperature and concentration profiles, an abrupt change in the concen-

tration profiles when the CO is depleted, and a linear concentration profile after

complete conversion of the CO. The orthogonal collocation on finite elements

procedure is then attempted to minimize the numerical difficulties associated

 -273-

with orthogonal collocation.

Finlayson ( 1980) used the OCFE procedure for several examples with

extremely sharp gradients. including diffusion and reaction in a porous catalyst

pellet and transient convective diffusion. His analysis of the orthogonal colloca-

tion method, the finite difference method, and the orthogonal collocation on

finite elements scheme for these and other chemical systems shows that the

"orthogonal collocation method is by far and away the best method" except for

systems with sharp profiles. The OCFE technique is then preferred. Finlayson

points out that detailed, comparative studies of even simple two-dimensional

problems using OCFE are rare, due to the complexity of the procedure. T.b.is sec-

tion provides the first complete OCFE analysis of a packed bed reactor along

with comparisons to simple orthogonal collocation results. This OCFE analysis is

performed on the two-dimensional system resulting after the radial orthogonal

collocation presented in Section 3.3.5.

The orthogonal collocation procedure uses a series of polynomials, each of

which is defined over the entire range 0 :S ~ :S 1, as a trial function for the axial

collocation. Complications wilh this global procedure arise in the presence of

steep gradients or abrupt changes in the solution profile. In such situations. it

may be advantageous to use trial functions that are defined over only part of

the region and piece together adjacent functions to provide an approximation

over the entire domain. Using such a procedure, smaller regions can be used

near the location of the steep gradient. The OCFE technique involves using

orthogonal collocation within each of these elements. A similar analysis can be

performed using the Galerkin method within each element. Howeve!' for non-

linear problems, it may be necessary to use quadrature formulas to evaluate the

integrals, resulting in lengthy calculations.

 -274-

Two major forms of the OCFE procedure are common and differ only in the

trial functions used. One uses the Lagrangian functions and adds conditions to

make the first derivatives continuous across the element boundaries, and the

other uses Hermite polynomials, which automatically have continuous first

derivatives between elements. Difficulties in the numerical integration of the

resulting system of equations occur using both types of trial functions, and per-

sonal preference must then dictate which is to be used. The final equations that

need to be integrated after application of the OCFE method in the axial dimen-

sion to the reactor equations (radial collocation is performed using simple

orthogonal collocation) can be expressed in the form

C da = Aa- f(a) (3.5-1)

dt

Using Hermite interpolation Within the elements, the C matrix is not diago-

nal. Explicit solution methods cannot be applied to this system of equations

easily because of this nondiagonal matrix, and most integration packages are

also not suitable for such a system. However using Lagrangian functions Within

the elements, the continuity and boundary conditions have no time derivatives,

and the C matrix in the above equation is diagonal with nonzero elements

representing a residual and zero diagonal elements for the algebraic conditions.

Most standard integration packages are also not suitable for mixed systems of

algebraic and ditlerential equations. However, our computer programs for the

standard orthogonal collocation procedure (Appendix 4) are written for mixed

systems. Thus the Lagrangian functions are used in this OCFE analysis.

The OCFE discretization analysis and computer programs developed in this

work allow general. specification of the number of finite elements and ~he degree

of the collocation functions Within each element. This is again an advancement

over prior analyses where the same degree of collocation was often used within
 -275 ~

each element. The entire domain is divided as shown in Figure 3.5-1. Within

each of NE elements, we apply orthogonal collocation as usual using Lagrangian

functions. The residual is evaluated at the internal collocation points. Using the

Lagrangian polynomials, we need to guarantee continuity of the first derivatives

or fluxes between elements. In our analysis, we use the continuity of the first

derivative. We thus append NE-1 conditions at the element boundaries. Then

the solution has continuous derivatives throughout the domain. With the addi-

tional two boundary conditions at ~ = 0 and at ~ = 1, there are a sufficient

number of conditions to solve for tho profile at all of the collocation points and

at the element boundaries.

For the k th element, we define the transformation

hJ; :::;; Z(Hl) - Z(.t) (3.5-2)

so that the variable u is between zero and one in the element. 1 Redefining the

differential equations in terms of the variable u allows the application of stan-

dard orthogonal collocation within each element. Figure 3.5-2 illustrates the

local numbering system within each element. If we then define NE as the

number of elements, Nk as the number of interior collocation points in element

k, and Mh as Nk + 2, the relation between the global numbering i ,j and the local

numbering I ,J is given by

i = [kJ~-1 (Ni + 1)
=1
l+I
(3.5-3)
Zi =Z(A:) + UJh.t

Then starting from the most general form of the dimensionless equations

after the one point radial collocation, we can apply the OCFE procedUre. As an

example, let us consider the energy equation for the solid:

1. Note that z in this expression is the normalized axial coordinate (equivalent to t).
 -276 ~

(3.5-4)

z=O

z=l

All temperatures in this equation are in terms of the conditions at the radial

collocation point.

Then for k = 1 to NE
Z = Z(.t:) + uh.t (3.5-5)

Thus within each element

with the overall boundary conditions

If we assume axial profiles within each element k = 1. ... , NE of the form

(3.5-7)

where as before 1.;, are lagrangian interpolation polynomials, then as usual

A dl;(u)
- _...;.....:._:....
du
j
ii

l
I.J-
(3.5-8)
B - d2l;(u)
I.J- duz ~

After applying orthogonal collocation.

 "277-

(3.3-9)

within element k for the interior collocation points i =2 .... , N.,t:+1.

Then if we require the continuity of the first derivatives between the ele-

ments so that the first derivative of the solution is continuous in the entire

domain 0 ~ t~ 1:

(3.5-10)

The system boundary conditions then only a*!ect the first and last elements:

ll
h ~ Al,J 9s1 = Agzs(911 1 -9g1)
1 .1=1
11 (3.5-11)
h NE
f
J'"'l
ANE.J 9a1 = Agzs(9g:w:
NE
-9SJ4 )
NE

This same analysis is then performed for the energy balance for the gas,

the energy balance for the thermal well, the two mass balances, and the con-

tinuity equation. The final coupled system of algebraic and differential equa-

tions is shown in Appendix 5, along with computer programs for steady state

solution using this OCFE procedure. It consists of 5N ordinary differential equa-

tions and N + 6(NE + 1) algebraic equations, where

(3.5-12)

The techniques used for the solution of this system are similar to those

used previously with the orthogonal collocation analysis. The majol'- complica-

tions are the potentially large number of resulting equations 2 and possible con-

z. Depending on the number of elements and number of interior collocation points per element.
 -278-

fusion and errors in setting up the indexing in the computer algorithm.

Extreme care must be taken in this last respect.

It should be pointed out that this approach using Lagrangian polynomials

gives identical results to those that would be obtained using Hermite polynomi-

als since on each element we use orthogonal polynomials of the same order,

since the boundary conditions are satisfied by both solutions, since the residu-

als are evaluated at the same points, and since the first derivatives are continu-

ous across the element boundaries. The only preference for one over the other

is for convenience. The Lagrangian formulation, however, has the added advan-

tage of being applicable in situations where the flux is continuous across the ele-

ment boundaries but the the first derivative is discontinuous. This can occur if

there is some type of physical change at the boundary.

A maJor advantage of the orthogonal collocation solution schemes is LhaL

the optimal location of the collocation points is automatically determined as the

zeros of the orthogonal polynomial after specification of the number of colloca-

tion points. This point carries through to the OCFE technique; however, in this

procedure, you must also specify the number and position of the elements.

Several means for this are available. The simplest involves preliminary solutions

using orthogonal collocation to provide reasonable estimates since, even though

the OC solution may be numerically unsatisfactory due to oscillations in the

profiles, it can give accurate indications of break points in the profiles.

If the OCFE solution procedure is to be used regularly for system analysis, a

careful study of the problem of locating elements is necessary. Finlayson ( 1980)

gives a basis for this study in a cursory examination of variable grid spacing and

elements sizes, or 'adaptive meshes' as he calls them. The simplest 'procedure is

based on physical information about the solution profile such as the point
 -279-

where CO is depleted within the reactor bed. Several direct mathematical

approaches are also discussed by Carey and Finlayson ( 1975) and Finlayson

( 1980). The first is based on the solution residuals. Although the residual is

zero at the collocation points when using orthogonal collocation or OCFE, it can

generally be nonzero elsewhere. After an approximate solution is calculated, the

residual is evaluated throughout the interval 0 :S;; ( :S;; 1. Ferguson and Finlayson

( 1972) and Carey and Finlayson ( 1975) show that the error in the solution is

bounded by the mean square residual. Since we want the residual to be small

everywhere, we can locate the elements to make the residuals approach zero

everywhere. In particular, additional elements should be inserted at points

where the residual is large. According to Carey and Finlayson ( 1975),

"in this way, as the number of elements is increased the

solution should converge faster than for a uniform spacing
of elements because the elements are placed where needed.
.. . However, the largest residual usually occurred at the
endpoint of the element since a continuity condition is
imposed there rather than setting a residual to zero. Thus
in other calculations it would probably suffice to calculate
the residual at the end points of the element and use that
residual to determine the location of additional elements."

Other mathematical procedures are described by Finlayson ( 1980) based on

extensions to Pearson's (1968) technique for finite difference methods and on

work by Ascher et al. ( 1979) and Russel and Christiansen ( 1978). These are not

discussed here since the above techniques are used in this thesiS.

3.5.2 Model Comparisons

A large number of steady state simulations were performed us~ this OCFE

procedure with various rate expressions to determine the applicability and

necessity of the technique for packed bed reactor modeling. Using the original

rate expressions of Lee (1973), Moe (1962), and Vatcha (1976), there is no major
 ~ 280-

necessity of the OCFE procedure since even for conditions With very steep

profiles, an accurate, non-oscillating solution can usually be obtained simply by

increasing the number of collocation points in the orthogonal collocation

analysis.:!! Due to the reversibility of the kinetics and the relatively low activation

energy. the profiles remain relatively smooth Without any abrupt changes or

sharp transitions.

The importance of the orthogonal collocation on finite element approach is

then considered using the single reaction non-reversible kinetics proposed by

Strand (1984). Figure 3.5-3 shows the steady state axial gas temperatures using

these kinetics With standard Type III (Table 3.4-5) operating conditions and With

the parameters shown in Tables 3.4~1 and 3.3-4 along With the exceptions Usg =
8.50, kow = 25.1, and Ea11 = 7000. Figure 3.5-4 shows a comparison of the steady

state axial temperature profiles using both orthogonal collocation (OC) and

orthogonal collocation of finite elements (OCFE) for the same conditions. As

shown. the numerical problems (oscillatory solution) exhibited in the orthogo-

nal collocation solution With six collocation points are completely eliminated by

increasing the number of collocation points or by using OCFE With appropriate

selection of the elements and order of collocation within the elements. The
major difference then between the OC solution with 12 interior collocation

points and the OCFE solution with three elements and only 10 interior colloca-

tion points (plus two interior element boundaries) is the solution time. The

OCFE steady state solution took almost exactly twice as long (123 seconds

versus 59 seconds)! Thus if at all possible, the orthogonal collocation procedure

is preferred, even at the expense of using additional interior collocation points.

Another major difficulty of the OCFE procedure is displayed in Figure 3.5-5.

3. Figures 3.4-26b and 3.4-20b show simulations 1lll'i.ng orthogonal collocation with possible numeri-
cal difficulties.
 -281-

This is the non-trivial problem of optimal selection of the number and position

of finite elements. Figure 3.5~5 shows a comparison of the OCFE steady state

axial temperature profiles for the conditions of Figure 3.5·3 with three elements

of length 0.15, 0.70, and 0.15 with N 1 =2, N =5, and N =3 and with three ele-
2 3

ments of length 0.10, 0.80, and 0.10 with N 1 = 1. N2 = 5, and Ns = 1. Although

these finite element schemes are only slightly different, the resulting solution is

significantly affected and is nearly as bad as the orthogonal collocation solution

with too few collocation points.

A final example shows the real power of the OCFE procedure. Figures 3.5-6

and 3.5-7 show 'best' solutions obtained using orthogonal collocation and OCFE

for the steady state axial temperature and concentration profiles for the non-

reversible kinetics where the reaction is extremely rapid With complete conver-

sion early in the reactor bed. These conditions are identical to those in Figures

3.5-:1 t.n 3 5-5 P.xcept that. the reaction rate is twice as large (k 011 =50.2). As we

would expect, the profiles are very sharp, and there is an abrupt change in the

concentration profiles at the point of complete conversion. The 'best' solutions

obtainable using orthogonal collocation require ten interior coUocation points

and still show significant numerical problems in both the temperature and con-

centration profiles. The use of more or less collocation points increases the

oscillations m the profile, espeCially m the vicinity or zero concentration.

The solutions using the OCFE procedure show dramatic differences in the

profiles. Based on an expected abrupt change in the concentration profile at

about ( =0.40 (from the preliminary orthogonal collocation results), an OCFE

solulion was obtained using two elements of length 0.4 and 0.6 with N1 = 5 and
N2 = 1 or a total of only six interior collocation points and one interior bound-

ary point. These OCFE axial profiles are much sharper than those predicted by
 ~ 282-

the orthogonal collocation procedure simply because orthogonal collocation

actually smooths the steep profiles by attempting to fit a single polynomial

through the entire region. Figures 3.5-6 and 3.5-7 show the locations of the col-

location points for both schemes. Grouping the collocation points around the

region of sharp profile changes in the OCFE procedure allows more accurate

representation of these changes. This is identical to the concept of increasing

the number of grid points around a sharp transition when using standard finite

element techniques.

AE shown in Figure 3.5-?, the OCFE procedure is also capable of handling

the abrupt concentration profile change where the reaction reaches completion

with only a single collocation point being necessary in the final element. while

the orthogonal collocation procedure shows significant error in the protlles even

with additional collocation points. Increasing the number of collocation points

in either element for the OCFE solution has no significant beneficial effect on the

axial profiles and can introduce oscillations in the first element as shown in Fig-

ure 3.5-8. Increasing the number of elements without using sufficient interior

collocation points or misplacing the elements can lead to very strange results

(Figure 3.5-9 and 3.5-10) due to the individual polynomial representations and

continuity conditions at the element boundaries.

Table 3.5-1 summarizes the steady state solution times and qualitative

behavior of the resulting axial profiles for a variety of solution schemes. Notice

that the solution times increase rapidly as the number or collocation points is

increased with both solution techniques. The numerical computation times for

the orthogonal collocation and OCFE solutions shown in Figures 3.5-6 and 3.5-7

are 11.3 minutes and 1.5 minutes, respectively, thus providing further evidence

of the necessity of the OCFE procedure ror these particular conditions.

 -283-

Based on these and other simulations along with those by Finlayson (1974),

we can conclude that. tor chemical packed bed reactor modeling, orthogonal

collocation is generally optimwn in terms of accuracy and numerical solution

times. However if steep gradients or abrupt profile changes occur within the

reactor bed due to extremely fast kinetics or complete conversion of the limit-

ing component, OCFE may be more accurate and faster than simple orthogonal

collocation. It is then recommended that, if such conditions are expected, the

modeling computer programs be written to allow either procedure to be used.

Otherwise, the programs can be written only for orthogonal collocation. The

computer programs in Appendix 5 are written for OCFE but allow orthogonal col-
location simulation simply by letting NE equal one.

Procedure hi N.,; Solution Time Anal Profiles

6 4:54 Ba.d.
7 5:28 Bad
oc 8 6:07 Bad
10 11:20 Bad
12 14:06 Bad
5, 1 1:33 Excellent
e. 1 1;18 Slight OscillaLion
7, 1 2:19 Slight Oscillation
5,2 4:51 Excellent
0.4, 0.6 5,3 6:27 Excellent
5,5 11:28 Excellent
OCFE 4, 3 5:44 Excellent
3, 3 4:19 Bad
0.15, 0.25, 0.60 1, 5. 1 2:15 Fair
2,5, 2 14:01 Fair
0.15, 0.20, 0.65 3, 5, 2 17:04 Bad
0,15, 0.15, 0.70 4 5,3 15:27 Bad

Table 3.5-1
OC and OCFE Solutions tor Conditions in Figure 3.5-8
 -284-

ELEMENT BOUNDARIES

/ \ \
:.-h·-:1. • • .c h·~
i=l 2 3 4~ 6 7 6
•
.'
Figure 3.1:>-1
Orthogonal Collocation on Finite Elements
Global Numbering System

/ELEMENT .t

I • • • • I
I= 1 2 3
u=O

Figure 3. 5-2
Orthogonal Collocation on Finite Elements
Local Numbering System
 -285-

650

t.::J
LLJ
0

600
LLJ
a:
::J
1-
a:
a:
LLJ 550
a_
:L
LLJ
1-

r.n
cr: 500
L::>

o.o 0.2 o.~ o.6 o.e 1.0

AXIAL POSITION, Z/L

Figure 3.5-3
Orthogonal Collocation Axial Temperature Profiles
Standard Type III (Table 3.4-5) Conditions
Non-Reversible Kinetics (Usg=B.5. k 0 m=25.1. Eell=7000)

- - N = 12 ----- N=B - - - N=6

 -286-

650
:X::
<..:)
lJ.J
0

600
lJ.J
a:
::::l
1-
a:
a:
lJ.J 550
0..
:r
lJ.J
1-
If)
a: 500
<..:)

0.0 0.2 0.11 0.6 O.B 1.0

AXIAL POSITION. Z/L

Figure 3.5-4
Axial Temperature Profiles
Same Conditions as Figure 3.5-3

OCFE solution with 3 elements of length O.lb, 0.70, 0.15

N1 = Z, N2 = 5, Ns = 3
- - - -- - OC solution with N = 12
 ~ 287-

650

600
IJ...J
a:
:=J
1-
a:
a:
IJ...J 550
a...
~
IJ...J
1-

(f)
a: 500
t.:J

0.0 0.2 O.Lt 0.6 0.8 1.0

AXIAL PClSITION. Z/L

Figure 3.5~5
Axial Temperature Profiles
Same Conditions as Figure 3.5-3

OCFE solution with 3 elements of length 0.15, 0.70, 0.15

=
N1 2, N2 5, Na 3= =
---- OCFE solution with 3 elements of length 0.10, 0.80, 0.10
N1 = 1. N2 = 5, N3 =1
 -288-

725
• Collocation Points
700
:X:
675
1..:)
I..L.J
a sso
U,.j 625
a:
::::>
1- 600
a:
a: 575
I..L.J
Q..
::£:
I..L.J
1-

(J")
525
a:
t.:)
500 Element Elemtflt 2

475
o.o 0.2 0.4 0.6 0.8 1.0
AXIAL POSITION. Z/L

(a)

725
• Collocation Points
700
:s::
1..:)
675
I..L.J
Cl 650

I..L.J 625
a:
::::l
1- 600
a:
a: 575
w
Q..
z: 550
w
1-

(J")
525
a:
t.:)
500

475
o.o 0.2 Q,lj 0.6 0.8 1.0
AXIAL POSITION. Z/L

(b)

Figure 3.5-6
Axial Temperature Profiles
Same Conditions as Figure 3.5~5 except Double the Reaction Rate (knu:=50.2)

a) OCFE wi.th 2 elements of length 0.4 and 0.6. N1 =5. N2 =1

b) OC with N=lO
 -289-

• Collocation Points

-
'a
0.5
IE---EIIImenl 1---+!~--- Eklmenl 2----~
X
0.1!
0
u
::z:
0
1-
u 0.2
a:
a:
LL.
o. l

0.0

-o.t~--~--L---L---L---L---~--~--~--L-~
0.0 0.2 0.1! 0.6 O.B 1.0
RXIRL POSITION. Z/L

(a)

0.6

o.s • Collocation Points

iCJ

·"' 0.1!
Cl
u
z 0.3
0

1-
u 0.2
a:
a::
lJ...

LW
0.1
-'
Cl
:J:
0.0 ----------- -
-0.1~-L---L---.L---~--L---L---L---~--~__J
0.0 0.2 0.4 0.6 0.6 1.0
AXIAL POSITION, Z/L

(b)

Figure 3.5-7
Axial Concentration Profiles
Same Conditions as Figure 3.5-6

a) OCP'E with 2 elements of length 0.4 and 0.6, N1 =5, N2 =1

b) OCwithN=lO
 -290-

725

700
X:
675
t.:l
w
0 650

w 625
a:
::::>
1- 600
c;;
a: 575
w
a_
:1:
llJ 550
1-
(f)
525
a:
t.:l sao --~..t.E----Etem.,..t 2----~

Y75
0.0 0.2 O.ij 0.6 0.8 1.0
AXIAL POSITION. Z/L

(a)

0.6

io 0.5
----li•:+-----Eiement 2 ----~
X
0.14
0
u
z 0.3
0
I-
u 0.2
a:
a:
u._

w 0. I
_J
0
:I:
0.0

-0.1
0.0 0.2 o.Y o.s o.e 1.0
AXIAL POSITION. l/L

(b)

Figure 3.5-8
Axial OCFE Temperature and Concentration Profiles
Two Elements of Length 0.4 and 0.6
Same Conditions as Figure 3.5-6

a. Gas Temperatures b. CO Concentrations

---- N1 =6. N2 = 1
 ~ 291-

• Collocation Points

-' o o.s El-111 t~Eitmtlftl 2 ~~<E------ Element 3-----~

I I
I I
0.4 I I
D 1 I
u I I
I

~ 0.3
1-
u 0.2
a:
a:
1.1....
w o. 1
.....J
D
:.E o.o

-0.1~--._--~ ___.__~----~--~--~--~-~--~
0.0 o.~ o.ij o.s 0.8 1.0
AXIAL POSITION, 2/L

Figure 3. 5-9
Axial OCFE Concentration Profiles
Same Conditions as Figure 3.5-6
hl=O.l5, h 2=0.20, hs=0.65; N1=3, N2=6, N3 =2
 -292-

750
725 • Collocation Points
X:
700
(..!)
LLJ 675
Cl

650
LI.J
a:: 625
=>
1-
a: 600
a::
LLJ
0.. 575
::1:
I
LI.J 550 I
1- I
I
U1 525 I
rr I
<..::l I
500 Elemtnt ~Element
I
E:lt-13
1
I
l!75
0.0 0.2 0.'-t 0.6 O.B 1.0
AXIAL POSITION. Z/L

F~ure 3.5-10
Axial OCFE Gas Temperature Profiles
Same Conditions as Figure 3.5-6
h 1=0.15, ~=0.25, hs=0.60; N1=l. N2=S, Ns=l
 -293-

3.8 CONTROL IIODEL DEVELOP'IIENT

As discussed by Ray (1981), the feedback control of distributed parameter

systems requires a reduction of the distributed system to an appropriate

lwnped parameter modeL Although this lumping can be performed prior to or

subsequent to a control structure analysis, the modeling equations must be

lumped for numerical integration during implementation. Our analysis of the

mathematical relationships describing the chemical and physical processes

within the reactor have used orthogonal collocation (OC) or orthogonal colloca-

tion on finite elements (OCFE) to replace the full distributed parameter system

with a lumped approximation that allows simulation of the steady state and

dynamic behavior of the packed bed catalytic reactor. Rather than making

major simplifications a. priori, the system analysis and numerical lumping used

a detailed mathematical description of the reactor. However, computing facili-

ties generally available for on-line control cannot currently perform the neces-

sary calculations rapidly enough for practical control applications with the full,

nonlinear model developed to this point. Furthermore, solution times for

dynamic simulations with this model even make detailed parameter studies and

process optimization impractical using the full model. Thus a simplified lower-

order model is desired for on·line multivariable control and for process studies.

3.6.1 State-Space Representation

In simplifying the process model. it is advantageous if the equations can be

reduced to fit into the framework of modern multivariable control theory, which

usually requires a model expressed as a set of linear first-order ordinary

difl'erential equations in the state-space form·

 -294-

~t) = JU[(t) + Ekl(t) + Ihi(t)

(3.6-1)
y(t) = Hx(t) + Fu(t)

where x(t), u(t), d(t), and y(t) are the state, control. disturbance, and measure-

ment vectors, respectively.

Since the orthogonal collocation or OCFE procedure reduces the original

model to a first-order ordinary differential equation system, linearization tech-

niques can then be applied to obtain the necessary representation. A major

advantage of the orthogonal collocation or OCFE procedure for the reduction of

the partial differential equations over previous techniques such as finite

differences or finite elements is that the size of the resulting ordinary

differential equation system is inherently quite small. Further reduction may be

desired and is discussed in Section 3.7. Once the dynamic equations have been

transformed to the standard state-space form and the model parameters are

estimated, various procedures can be used to design one or more multivariable

control schemes, or the reduced model can be used for detailed process simula-

tion and optimization.

However. real processes cannot accurately be described by the determinis-

tic form of Equation (3.6-1). For control applications, modeling and measure-

ment errors, reduction inaccuracies, and noise in the system may need to be

accounted for by considering the addition of stochastic disturbance terms to

the equations:

i(t) = JU[(t) + lkl(t) + Ihi(t) + l{t)

y{t) = Hx(t) + Fu(t) + f](t) (3.6-2)

where 1(t) is a vector of random process noise and 7'](t) is a vector or random

measurement error. These can be thought of as representing the total effect of

all of the disturbances or noises in the process not accounted for by the deter-

ministi~ model. These stochastic disturbances can include uncontrolled or

 p 295-

unmodeled fluctuations in inlet concentrations or flow rates, in catalyst activity,

in cooling wall temperature, and in heat transfer coefficients and can lead to

suboptimal reactor performance or even system uncontrollability.

In developing a general state-space representation of the reactor, all possi-

ble control and expected disturbance variables need to be identified. In the fol-

lowing analysis, we will treat the control and disturbance variables identically to

develop a model of the form

i(t) =AI{t) + Ww(t) (3.6-3)

y(t) =Hx(t) + Fw{t)
where w{t) now contains all control and disturbance terms,

Ww(t) = Bu(t) + Dd(t). This model can then be used for process simulations.

When specific control studies are desired, the Ww(t) term can simply be

separated into the appropriate control term Bu(t) and disturbance term Dd(t).

depending on the selected control configuration. Examples of this are shown in

Section 4.3.

Consider lhe simplified tlow diagram of the reactor system shown in Figure

3.6-1. The possible control variables or disturbances to the process are the

tlowrates of the input gases, the recycle ratio, the cooling jacket temperature,

the heat load of the preheater, and the feed-effluent heat exchanger bypass.

From the reactor's point of view (and thus the reactor model), the inlet gas tern-

perature, the cooling wall temperature, the gas velocity, and the inlet gas con-

centration can be affected. We will thus consider the control and disturbance

vector of the form

(3.6-4)

where these variables are nondimensional and the superscript (0 ) or subscript

(0) indicates inlet conditions.

 ~ 296 ~

3.6.2 .llodel I..i.nearisation

The major complication in the model after reduction of the original partial

ditierential equations to ordinary ditrerential equations is the nonlinearities that

result from the nonlinear rate expressions and from the temperature, concen-

t:ration, and velocity dependencies of the convective terms and of the gas and

reaction properties. The reduction of the first-order ordinary differential equa-

tion model to the state-space representation involves linearization of the alge-

braic and differential equations about the steady state. The resulting
mathematical description is then valid for small perturbations around this

steady state.

This analytic linearization, although conceptually quite simple, is extremely

tedious. Simple numerical linearization could be used during simulations and

control through various readily available computer programs but does not pro-

vide a detailed analytical solution from which the effects of various parameters

can be investigated.

Minor reduction of the nonlinear ordinary differential equation model

developed using orthogonal collocation or OCFE first involved assuming constant

heat capacity of the gas throughout the reactor bed 1 and neglecting the axial

and radial mass diffusion terms, based on the results of the model analysis

presented in Section 3.4. The resulting model was then linearized in three

stages: linearization of the reaction rates, linearization of the algebraic equa-

tions and substitution into the differential equations, and linearization of the

resulting ordinary differential equations.

1. Simulations showed no loss in accUiacy by this assumption. Its only real benefit is a minor
:reduction in linearization etfort.
 -297-

3. 6. 2.1 Lineariza.titm. of the Reaction Rates

The nondimensionalized reaction rate expressions

(3.6-5)

(3.6-6}

(3.6-7)

v= Kq. YCH..YHeO (3.6-B)

KPJ'~P~ Yeo (YH8 ) 3

KP, = e l
Kp +
it
i =S, M (3.6-9)

(3.6-10)

are first linearized about the steady state operating conditions (ss) using a Tay-

lor series linearization approach. If we neglect the second and higher order
terms in the Taylor series expansion, the linearization is of the form

(3.6-11)

After lengthy analytic manipulation of the expressions, the resulting linearized

rate expressions are of the form given in the equation above with the [ BR
OZ
lA

terms defined in Appendix 6.

It is important to remember that the linearized expressions are only valid

 -298-

in the Vicinity of the specified steady state conditions. A comparison of the rela·

tive magnitudes of the linear and nonlinear reaction rates is presented in Table

3.6-1 for both Type I and Type II standard operating conditions defined in Table

3.4-5. These results are presented mainly to show the necessity for caution in

using the linearized expressions since significant errors can occur away from

the steady state. In general, although the rate expressions are highly nonlinear,

the errors are relatively small even at considerable deviations from steady state.
Although the errors in the steam-shift rate are quite large in several cases, the

reaction rates in these cases are so small that the steam-shift reaction has

insignificant effect on the reactor dynamics.

Type I Conditions

%Error
T. Tg Yt Y2 Rv Rs
steady state 630 627 0.13 1.00
640 637 0.13 1.00 0.23 12.5
650 647 0.13 1.00 0.85 25.1
620 617 0.13 1.00 0.27 24.2
630 627 0.25 0.99 0.27 7.5
630 627 0.03 1.02 0.91 4.2
64-o 637 0.25 0.99 4-.03 2.5

Type n Conditions

%Error
T. Ta Yt Y2
Ru Rs
steaty state 613 610 0.71 1.01
603 600 0.71 1.01 0.31 0.57
623 620 0.71 0.71 0.26 0.45
613 610 0.82 1.01 0.15 2.60

Table 3.6-1
Nonlinear vs. Linearized Reaction Rates

3.6.2.2 Linearization of the Algebraic Equa.tions

The algebraic equations for the orthogonal collocation model (shown in Sec-

tion 3.3.5) consist of the axial boundary conditions along with the continuity
 -299-

equation solved at the interior collocation points and at the end of the bed. This

latter equation is algebraic since the time derivative for the gas temperature

can be replaced with the algebraic expression obtained from the energy balance

for the gas.

The boundary conditions for the mass balances and for the energy equation

for the thermal well can be solved explicitly for the concentrations and thermal

well temperatures at the axial boundary points as linear expressions of the con-

ditions at the interior collocation points. The set of four boundary conditions

for the gas and catalyst temperatures are coupled and are nonlinear due to the

convective term in the inlet boundary condition for the gas phase. After a Tay-

lor series linearization of this term around the steady state inlet gas tempera-

ture, gas velocity, and inlet concentrations, the system of four equations is

solved for the gas and catalyst temperatures at the boundary points.

Linearization and solution of the continuity equation for the velocities at

the interior collocation points and at the end of the reactor are somewhat more

complex. Unfortunately, the time derivative of the gas temperature term adds

significant complexity to these algebraic equations. Since the linearized model

is expected to be valid over only a limited region around the steady state, and

since a sludy or the magnitudes of the terms in this equation showed that this

time derivative term is relatively small near the steady state, the term was

neglected (equivalent to assuming ":t' = 0). Simulations using the full, non-

linear model verified that negligible error resulted from dropping this term. The

maximwn error seen was during the first 10 seconds of a dynamic simulation

and was generally less that 0.1% even during this period.

The resulting equation

 -300-

(3.6-12)

for i = 1, ... , N+ 1 was then linearized using the folloWing relationship

(3.6-13}

The algebraic equations for the endpoint gas temperatures were then substi-

luted into the linearized continuity equations. These equations were then solved

for the velocities. The results of this linearization of the algebraic equations is

shown in Appendix 6.

3. 6. 2. 3 Linsarizatian of the IJiflerential Equatians

The linear reaction rate expressions and the linear expressions for the

velocities and for the concentrations and temperatures at the axial boundary

points were then substituted into the ordinary differential equations. Lengthy

algebraic manipulation waS' necessary to linearize these equations due to

extreme couplings between the different state variables. The linearization was

completely analytical and retained all possible complications. The results of

this analysis are shown in Appendix 6. Again great care must be exercised dur-

ing this algebraic reduction due to the extreme complexity of the model. All

results were verified through careful model simulations and ftnally by using the

symbolic equation solver SMP available on the Caltech computer system.

The final linearized model is of the form

i: = Ax + lfw + C (3.6-14)

where the state vector

 -301-

consists or the catalyst temperatures, the gas temperatures. the thermal well

temperatures, the CO concentrations, and the C02 concentrations at the interior

collocation points. The control and disturbance variables are

(3.6-16)

This linear system is then of order 5N, where N is the number of interior colloca-
tion points. Analytical expressions for the elements of the matrices 'A and W are

shown in Appendix 6. The constants C in the equations can be easily eliminated

by using deviation variables about the steady state values. At steady state, i: = 0

and thus

O=Ai:+fi+C

Subtra.cting this from the original equation,

:i: =A(x-x) + W(w-w) (3.6-18)

Letting

I'=x-x and w' = "--w (3.6-19)

the resulting equation is of the desired state-space representation:

r =AX+ lfw' (3.6-20)

Note that in the remaining analyses, we will drop the ('). It should still be clear

that deviation variables are being used. Then as per the previous discussion,

this linear representation can easily be separated into the standard state-space

representation shown in Equation (3.6-1) for any particular control

configuration.
 -302-

3.6.2.4 Numerical Solution of tha Li:nearized. Model

Numerical simulation of the behavior of the reactor using this linearized

model is significantly simpler. The first step in the solution must be to solve the

full, nonlinear model for the steady state profiles. This is accomplished using

either the algorithm by Powell (Rabinowitz, 1970) or Brown ( 1967) discussed in

Section 3.3.5. This first step, is limiting in the sense that no further linearized

analysis is possible without it. Due to the complexity of the nonlinear model.

numerical convergence problems may be significant during this steady state

solution. Obviously. the success of this solution is often based on the initial

guesses for the steady state profile. Experience or empirical data can help in

setting these values. Additionally, under extreme cases, the nonlinear dynamic
simulation programs can be used to simulate the behavior of the system from

specified profiles to the final steady state. As the dynamic solution approaches

the steady state, the resulting profiles can be used as the initial guesses for the

steady state solver so that the dynamic solution does not need to continue com-

pletely to steady state.

The steady state profiles are then used to calculate the matrices A and Y.

The Adams-Moulton predictor-corrector technique can again be used t.o numeri-

cally integrate the linear model for dynamic simulations. However, due to the

linearity of the resulting mathematical system, an analytical solution of the

difi'erential equations is possible. The general solution of the clifl'erential system

given by Equation (3.6-20) is

(3.6-21)

Then if we consider a time interval (t -.. t + dt) over which w is constant, 2 this

solution simplifies to

2. We can take very small time steps, dt, i! needed.

 -303-

(3.6-22)

The major complications in this analytical solution are obviously the matrix

exponential and matrix inverse. If we let

A= SA s- 1 (3.6-23)

where A is a diagonal matrix of the eigenvalues of A and S is the corresponding

matrix of eigenvectors, we then get

(3.6-24)

where eA dt and A-t are easily calculated smce A is diagonal. Thus after the

steady state is determined and the time steps (dt) are selected, we can calculate

the eigenvalues and eigenvectors of the matrix A and from these the values of
A- 1 and eAdt. These lengthy calculations only need to be performed once for

each steady state and selected time step! Of course these time steps can be

quite large with the only restriction being that the control and disturbance vari-

ables are nearly constant during this period. If these variables are constant
over more than one selected time step, even the value of A- 1(eAdt_I)Ww' will be

constant and will not have to be recalculated between time steps.

Several computer programs have been written to simulate this linearized

model of the reactor. These programs are discussed and listed in Appendix 4.

Basically, two programs are usedfor the linear model. The first called LINMOD

uses the Adams-Moulton predictor-corrector numerical integration of the linear

differential equation system, including our variable time-step analysis. This pro-

gram is similar in structure and operation to the nonlinear program NLNMOD.

The second program written for the linear model simulation. ANAMOD. uses the

analytical solution· of the linear equations rather than performing a numerical

solution. Thus it takes only a fraction or the solution time.

 -304-

These programs are set up to automatically determine the steady state

solution, to calculate the coefficient matrices for the linearized modeL and to

optimally perform the dynamic simulations for a large variety of step changes

or disturbances or for start-up conditions. The program ANAMOD only performs

the matrix inversions, exponentiations, and multiplications as necessary. The

matrix inversion or the eigenvector matrix simply involves transfornung the

matrix into upper triangular form by successively multiplying it on the left by a

transformation matrix that preserves the determinant of the original matrix.

The triangular matrix is inverted by back substitution and ftnally the inverse of

the original matrix is obtained by multiplying the inverse of the triangular

matrix by the original transformations on the right.

The determination of eigenvalues and eigenvectors of the matrix A is based

on a routine by Grad and Brebner (1968). The matrix is ttrst scaled by a

sequence of similarity transformations and then normalized to have the Eucli-

dian norm equal to one. The matrix is reduced to an upper Hessenberg form by

Householder's method. Then the QR double step iterative process is performed

on the Hessenberg matrix to compute the eigenvalues. The eigenvectors are

obtained by inverse iteration.

3.6.3 State-Space 'Simulatiom

Figure 3.6-2 shows the simulated dynamic behavior of the gas temperatures

at various axial locations in the bed using both the linear and nonlinear simula-
tion programs for a step change in the inlet CO concentration from a mole frac-

tion of 0.06 to 0.07 and in the inlet gas temperature from 57:3° K to 593" K. Fig-

ure 3.6-3 shows the corresponding dynamic behavior of the CO arid CO:a concen-

trations at the reactor exit and at a point early in the reactor bed. The axial
 -305-

concentration profiles at the initial conditions and at the final steady state

using both the linear and nonlinear simulations are shown in Figure 3.6-4, and

the axial gas temperature profiles are shown in Figure 3.6-5 at t = 0, 15 seconds,

and 100 seconds (steady state). The temporal behavior of the profiles shows

that the discrepancies between the linear and nonlinear results increase as the

final steady state is approached. Even so, there are only slight discrepancies

(less than 2% in concentrations and less than 0.5% in temperatures) in the

profiles throughout the dynamic responses and at the final steady state even for
this relatively major step input change.

Figure 3.6-6 shows dynamic linear and nonlinear simulations of the gas
temperatures at t = 0.17 and ( = 0.38 and Figure 3.6-7 shows the exit CO and
C0 2 concentrations for a 50 second disturbance in the inlet gas temperature

from 573° K to 601.65° K (a 5% increase) and in the inlet CO concentration from

a mole fraction of 0.06 to 0.072 (a 20% increase). Again we find that only minor

differences (less than 2%) are apparent between the two models and only then at

conditions sufficiently far from the steady state. As expected, the models give
identical results when the orl.ginal steady state is reestablished after the distur-

bance.

Figure 3.6-8 shows the temporal behavior of the exit gas temperature and

the gas temperature at ( =0.38 under start-up operation using both the linear
and nonlinear simulation programs. Figure 3.6-9 shows the axial temperature

profiles at 15 seconds, 30 seconds. and at steady state. As expected, the major

deviations between the models occur early in the simulation where the reactor

conditions are relatively far from the steady state conditions around which the

model is linearized. However, the magnitudes of the deviations are actually very

small, thus verifying the ability of the linearized model to simulate reactor

behavior even relatively far from steady state.

 -306-

Thus the simulations show that the linear model is quite accurate even for

relatively large deviations from the steady state. Of course, losses in accuracy

are greater the larger the deviation. For start-up and disturbance simulations,

the linearized model does predict an eventual return to the steady state around

which the system was linearized. However for step input changes where the final

steady state differed from the original, some minimal loss in accuracy is

apparent in the final steady state reached using dynamic simulations of the

linear model from the original steady state. This difficulty can easily be circum-

vented in the case of step changes by relinearizic.g about the new final steady

state conditions somewhere during the simulation.

Table 3.6-2 shows a comparison of the solution times for the various

dynamic simulations using the three models. As shown, the reduction in solu-

tion time is enormous using the linear modeL Using numerical integration of

the linear state-space model, simulation times are reduced by a factor of 10.

The analytical solution reduces simulation times by another factor of up to 100

(to a point where real time .solution is possible)! As expected, the results using

both the numerical and analytic solutions of the linear model are identicaL

llodel
Simulation
NLNMOD LINMOD ANAMOD
T0 ... 593cK
Step 2:54:24 12:54 00:36
x:m--0.07
Step T., ... 593°K 3:14:41 18:04 00:34
Start-up 4:43:26 28:39 00:29
Disturbance T0 -+ 60::~°K 23:10
4:58:42 00:31
(50 seconds) x:m .... 0.072
Table 3.6-2
Comparison of Simulation Times

Thus the results of this investigation concerning locally linearized models

along with the results of prior investigators (Heiberg et al., 1971; Sinai and Foss,
 -307-

1970) show that such models are quite descriptive of the overall system and are

an extremely useful means of reducing model complexity. AB pointed out by

Hoi berg et al. ( 1971) and verified in this analysis, there is relatively little incen-

tive for retaining the nonlinear representation in View of the large uncertainties

in the numerical values of the physical and chemical parameters.

 -308-

BYPASS
co---.
H2
N2
CH-4---'

..,..._
PRE HEATER COOLING
OIL

FEED-EFFLUENT
HEAT EXCHANGER

Figure 3. 6-1
Simplified Reactor Flow Diagram
 -309-

630

z= 0.38L
y:

CJ
620
-----
w
0

w
a:
610
------
--------
z =l --------

:::)
1-
cr:
a: 600
LU
a..
:E
LU
1-

(f)
590
a:
<..:)

sao~~--~--~~--~--~~~~----~--~~----~
0 IS 30 60
TIME !SECJ

Figure 3.6-2
Gas Temperature During Step Input Change
Standard Type II (Table 3.4-5) Conditions
(To from 573° K to 593° K. x& from 0.06 to 0.07)

- - Nonlinear Model - - - - - linear Model

 ~ 310 ~

0.8 1. 65

- 0.7
N
'o
'o
- 1. 60
~

v----
X
~
CO, z = 0.17L
(\J
D
u 0.6 C0 2 , z =L D
u
z 1. 55
D ~ z
D
....... 0.5 .......
1- C0 2 • z ,. 0. 17l 1--
u u
a: 1. so a:
a:
Ll...

w
.....1
v
0.4: r-
F-
=--
---- CO , z ,. L a:
Ll...

w
1. 45 .....1
D D
:t: 0.3 f-
:s:

0.2 1. 4:0
0 15 30 60
TIME (SECl

Figure 3.6-3
Concentrations During Step Input Change
Standard Type II (Table 3.4-5) Conditions
(To from 573° K to 593° K. x& from 0.06 to 0.07)

- - Nonlinear Model --- -- linear Model

 - 311 -

0.08 1. 60

N
'o
.....
X
0 1. 55
u (\J

z 0.06 0
u
-
0
1-
u 1. 50
z
0
.......
cr: I-
a:: u
lJ.... a:
w a::
O.Oll lJ....
_J
0 l.llS w
:L _J
0
::s::

n.n2 1. 40
o.o 0.2 0.4 0.6 0.8 1.0
RX1RL POSITION, Z/L

Figure 3. 6-4
Axial Steady State Concentrations Before and After Step Change
Standard Type 1I (Table 3.4-5) Conditions
(To from 573° K to 593° K. x80 from 0.06 to 0.07)

- - Nonlinear Model - - - - - linear Model

 -312-

650

~
630
t:)
w
0
---"
w 610
a:
:::>
I-
cr:
a::
w 590
a_
::E
w
I-
(f)
570
cr:
t:)

550
o.o 0.2 0.1! 0.6 0.8 1.0
'"AX I RL POSITICJN, Z/L

Figure 3.6-5
Axial Gas Temperaure Profiles After Step Change
Standard Type II (Table 3.4-5) Conditions
(To from 573c K to 593c K. x& from 0.06 to 0.07)

- Nonlinear Model - - - - - linear Model

 ~ 313-

650

:X::
630
<..::>
w
D

lJ.J 610 z "' 0.3BL

a:
::::J
1-
a: z= 0.17L
a:
w 590
Cl...
L
w
1-
570
c.n
a:
t:)

550
0 30 60 90 120 150
f

TIME lSECl

Figure 3. 6-S
Gas Temperatures During 50 Second Disturbance
Standard Type II (Table 3.4-5) Conditions
(To from 573° K to 602° K. x& from 0.06 to 0.072)

- - Nonlinear Model ----- linear Model

 - 314-

0.6 1.6
_______ ,
\ "''o
'o \ _..
0.5 \ C0 2 X
X 1.5
f\.1
D
u D
u
z z
-
D
1-
u
Q,ij l.ij
D
......
1-
u
a: co a:
0:: 0::
u.. 1.1....
w 0.3 1. 3 I.J..J
-'
D
....J
D
:L X:

0.2 1.2
0 30 60 90 120 150
TIME CSECl

Figure 3.6-7
Exit Concentrations During 50 Second Disturbance
Standard Type II (Table 3.4-5) Conditions
(To from 573° K to 602° K. x& from 0.06 to 0.072)

- - Nonlinear Model ----- linear Model

 -315-

630

:3:::
t:)
610 z =l
w
0
z "' 0.38L
w
a:
:::>
1- 590
CI
a:
w
a_
~
w
1- 570
(J")
a:
t:)

550
0 ao ijQ 60 80 100
TIME (SECl

Figure 3. 6-8
Temporal Behavior of Gas Temperatures during Start-Up
Standard Type II (Table 3.4-5) Conditions

- - Nonlinear Model - - - - - linear Model

 -316-

630

:::r:::::
Ste11dy state
Cl
w 610
0

w
cr:
::::>
....... 590
a:
cr:
w
o_
~
w
....... 570
(.()
a:
t.:)

550
o.o 0.2 0.11 0.6 0.6 1.0
AXIAL PClSITIClN, Z/L

Figure 3.6-9
Axial Gas Temperature Profiles during Start-Up
Standard Type 11 (Table 3.4-5) Conditions

- Nonlinear Model - - - - - linear Model

 -317-

3.7 MODEL DIMENSIONALITY

Mathematical models are widely used in the design. simulation. optimiza-

tion, and control of chemical packed bed reactors. Although low-order models

are often developed for control of simple chemical systems using statistical

analysis of input/output data, these empirical models are generally inappropri-

ate for complicated, highly nonlinear packed bed analyses where very low-order

models are not sufficient and where operating conditions may vary greatly from

the vicinity of the operating point for which the models were developed. For

these systems. the mechanistic modeling approach presented in this thesis is

necessary. In this mechanistic analysis, a detailed mathematical description of

the physical and chemical phenomena in the system is solved directly. Although
this generally provides a more detailed solution valid over a wider operating

range and allows detailed investigation and prediction of the reactor behavior,
the model even after linearization is often too complex for controller design or
for implementation as part of the actual control system. Low-order models are

thus required for on-line implementation of multivariable control strategies.

This section presents a dimensionality study of the model discretization

and physical modeling assumptions for the mechanistic packed bed reactor

model. A further discussion of model reduction to approximate the full system

with one of moderate order (2Nth-order) is also presented. It is our opinion that

further reduction probably is not needed or desired for control studies. lf addi-

tional reduction is necessary, model reduction techniques such as those

presented by Bonvin (1980) and Wilson et al. (1974) should be attempted.

3.7.1 Model ~tion

The model discretization or the number of collocation points necessary tor

 -316-

accurate representation of the profiles within the reactor bed has a major effect

on the dimensionality and thus the solution time of the resulting model. As pre-

viously discussed, radial collocation with one interior collocation point generally

adequately accounts for radial thermal gradients without increasing the dimen-

sionality of the mathematical system. However, multipoint radial collocation

may be necessary to describe radial concentration profiles. The analysis of Sec-

tion 3.4.5 shows that even with very high radial mass Peclet numbers, the radial
concentration is nearly uniform and that the axial bulk concentration and

radial and axial temperatures are nearly unaffected by assuming uniform radial
concentration. Thus model dimensionality can be kept to a minimum by also

performing the radial concentration collocation with one interior collocation

point.

In this section, the optimal choice for the number of axial collocation

points is discussed. Obviously, if the number of axial collocation points is

insufficient. the resulting axial solution for the temperature and concentration

profiles will be incorrect. On the other hand, one of the problems with fitting
high-order polynomials to the axial profiles is that the polynomials, if of

sufficiently high order, may begin to ripple along the curve. As pointed out by

Jutan et al. ( 1977), this rippling can be extremely detrimental since the colloca-

tion formulas are used to approximate derivatives. Since there is a tradeoff

between lowering the number of collocation points to reduce model dimen-

sionality and reduce profile rippling and increasing the number of collocation

points to retam high simulation accuracy, extreme care is required in the selec-

tion of the number of axial collocation points.

The first attempts at determining the optimal number of axial collocation

points simply involved repeating simulations with differing axial discretization to

find conditions with minimum ripple and maximum accuracy. Figure 3.7-1
 -319-

shows the steady state axial temperature profiles for standard Type I (Table 3.4-

5) conditions as the number of axial collocation points is changed. Eight or

more axial collocation points provide similar results, and even simulations With

six coUocati.on points show minimal inaccuracy. However, reducing the number

of collocation points below this leads to major discrepancies in the axial profiles.

These errors are even more important in the dynamic simulations shown in Fig-

ure 3.7-2.

Although this simulation procedure provides an optimal choice of the

number of axial collocation points for steady state and can give an indication of

the appropriate number of collocation points for dynamic simulations, Bonvin

(1980) shows that the convergence pattern of the dominant eigenvalues of the

model as the number of collocation points LS increased can also be used as an

important procedure for determining the optimal axial discretization. Since

these dominant modes only describe the dynamic behavior of the linearized sys-

tem, simulation of the full model should be used to verify the results for tran-

sient and steady state analyses.

If the axial discretization is fine enough, the dominant eigenvalues have

converged to their true values. Figures 3.7-3 and 3.7-4 show the convergence of

the dominant eigenvalues for the Type I and Type II (Table 3.4-5) operating con-

ditions. The conditions for Figure 3.7-3 are the same as those used for Figures

3.7-1 and 3.7-2. The eigenvalue analysis shows that less than six collocation

points will lead to significant errors in the dynamic simulations tor these condi-

tions since the dominant eigenvalues are far from their true values. Seven col~

location points would seem lo be sufficient and six may be satisfactory. These

results verify those obtained above using simulations with differing discretiza-

tion (Figure 3.7-1 and 3.7-2). With six interior axial collocation points, the full

model would consist of 30 coupled, nonlinear ordinary differential equations

 -320-

along with their algebraic boundary conditions. The linear model would be a
30th-order state-space representation!

Under conditions where very steep gradients exist within the reactor bed,

collocation may lead to oscillatory axial profiles (rippling) due to the attempt to

fit an Nth-order smooth polynomial function to the axial behavior. Although

these oscillations can be often reduced by decreasing the number of collocation

points, significant errors in the profiles can then result. Simple trial·and-error

should be used to determine the best number of aXial collocation points to

reduce rippling and retain accuracy. In extreme cases with very steep axial

profiles or abrupt changes m the profiles. the orthogonal collocation on finite

element procedure explained in Section 3.5 may be necessary. Using this tech-

nique, not only the number of collocation points in each element needs to be
specified but also the number and size of the elements.

3.7.2 Ph}Sicalllodeling Simplifications

Due to the complexities of the full mathematical description of a packed

bed reactor and the complications with numerical solution and analyses of such

descriptions, the extensive use of physical modeling simplifications is common.

Assumptions that reduce the complexity of the resulting mathematical model

such as neglecting dispersion effects and radial gradients have been discussed

elsewhere in this work. Basically, the common assumptions of neglecting radial

temperature gradients, neglecting thermal dispersion. and neglecting the varia-

tion of physical properties significantly reduce the accuracy of the resulting

model for numerical simulations and for further optimization or control studies

without reducing the dimensionality of the resulting representation. In view of

the advances in computational techniques over the past decade for the solution
 -321-

of partial differential equation systems, 1 these simplifications are unnecessary

and provide minimal, if any, reduction in solution etlort. Transient reactor

behavior can thus be simulated using accurate three-dimensional modeling With

the inclusion of known dispersion effects and parameter variations.

To this point in the modeling analysis, we have tried to incorporate all of

the mechanisms necessary for accurate description of the physical and chemi-
cal phenomena occurring in industrial reactors and to study the ef!ects of vari-

ous modeling simplifications. We continue this study by considering several

major simplifications which do not simply reduce the structure of the partial
ditlerential equations but actually significantly reduce the number or necessary

equations or the dimensionality of the process. These include the extensive use
of pseudo-homogeneous models-those that do not distinguish between the con-

ditions within Lhe fluid and those on the solid catalyst-and the assumptions of

quasi steady state for concentrations and negligible energy accumulation in the

gas phase.

3. 7.2.1 Homogrmaous Analysis

The mathematical models of the heat and mass transfer processes in

packed bed reactors are classified in two broad categories:

1. One-phase or pseudo-homogeneous models in which the reactor bed is

approximated by a quasi-homogeneous medium.

2. Two-phase or heterogeneous models where the catalyst and ftuid

phases and the heat and mass transfer between phases are treated

explicitly.

1. In partiC'J.lar, the collocation techniques.

 -322-

Although the heterogeneous models are more realistic, they have been generally

ignored until only recently due to difficulties with numerical solution and

increased model dimensionality. Even several nonadiabalic packed bed reactor

studies that begin with a two-phase analysis reduce the model to a pseudo-

homogeneous form by assuming equal gas and catalyst temperatures for steady
state and dynamic solution (Jutan et al., 1977) or at least for the steady state

solution (Heiberg et al., 1971). Some authors (Yortmeyer and Schaefer, 1974;
Vortmeyer et al., 1974) have even considered the equivalence of the one- and
two-phase approaches for one-dimensional or adiabatic studies. This thesis

presents the first detailed two-dimensional heterogeneous dynamic and steady

state packed bed reactor analysis.

However since it has been common practice to develop models for catalytic

reactors that do not distingUish between thermal conditions within the tluid

phase and those on the solid catalyst, a cursory examination of the homogene-

ous analysis is presented here. This assumption is generally justified by the

expectations of small temperature differences between the solid and gas phases

and results in a one-phase ctintinuum representation of the actual reactor bed.

As discussed by Jutan et al. (1977). the difference between gas and catalyst tem-

peratures for fast flowing gas systems should be negligible. They further refer-

ence the work by Shaw ( 1974) that shows that there is ample driving force to

remove the heat generated by reaction and that no temperature difference

should exist between the gas and catalyst at steady state.

In our original system of partial differential equations, the two energy bal-

ances can be combined as proposed by Jutan et al. ( 1977) by eliminating the

u .
term, ~(T8 -Tg). that describes the heat transfer between the solid and gas. If

the gas and solid temperature are assumed equal (T1 =Tg). and the homogene-
 -323-

ous gas/solid temperature is defined as T, the original combined energy balance

for the gas and solid becomes

+ ( -AH:u:)Ru: + ( -AHs)Rs (3.7-1)

where kz and kr are effective conductivities for the combined gas/solid system.

Additionally, since

(3.7-2)

the coefficient of the time deri.vative reduces to

(3.7-3)

The boundary conditions are

6T
z=O k,. oz = UgCp,fgE (T -T0)

z=L aT = o
l'z
(3.7-4)
r=Ro kr oT = ht(T -Tt)
ar
r=R 1 -kr aT = hw(T -T,)
6r

where ht and hw are effective heat transfer coefficients between the gas/solid

medium and the thermal well and cooling waLL respectively.

This equation is solved with the continuity equation, the energy balance tor

the thermal well, 2 and the concentration equations using the same procedure as

for the full system. The dimensionality of the model has now been reduced by N

(i.e., the model now has order 4N). Computer simulation programs were written

2. With a."'l overall heat transfer coefficient being used rather than Uta and Utg.
 -324-

for this homogeneous model. Unfortunately, no direct comparisons can be

made between the simulations with the two-phase analysis and the homogene-

ous analysis since no direct mathematical relationship can be made between the

effective and overall heat transfer coefficients of the homogeneous model

(kz, kr, ht. hw, Ut} and the individual gas and solid coefficients of the two-phase

model (kzs• kzg, krs, krg. hts. hq, ht.g. hwg• Uts. Ut.g). Some attempts (Vortmeyer and
Schaefer, 1974) have been made in this regard, but are not accurate enough for

direct comparison of the simulations of the two models.

Solution times using the homogeneous model are 15% to 25% less than that
for the full two-phase analysis, although the accuracy of the results may be
somewhat in question. Figures 3.7~5 and 3.7-6 show the axial temperature and

concentration profiles with Type 1 and II (Table 3.4-5) operating conditions and
with realistic heat transfer parameters. Although these simulations appear

similar to those obtained with the heterogeneous analysis, no direct comparison

is possible as explained above.

However, simulations usi,p.g the heterogeneous analysis even with high inlet

velocities lead to steady state temperature profiles with differences of up to 10°

K between the solid and gas phases and transient results with differences up to

20" K due to the high exothermicity of the methanation reaction on the catalyst

surface. These temperature differences can lead to significant errors in the

predicted reaction rates. Using the kinetics proposed by Strand (1984), the

methanation reaction rate about doubles with ever 10° K temperature rise over

the normal operating region. Therefore although a homogeneous analysis may

be adequate for reactions that are only slightly exothermic or simulations of

steady state phen:omena, a heterogeneous analysis is necessary for dynamic

simulations of highly exothermic reaction processes.

 -325-

The increased dimensionality of the heterogeneous system can be reduced

using other techniques discussed in this section which require much milder

assumptions. The assumption of negligible energy accumulation in the gas

phase based on the approximation (Equation 3. 7-3) provides the same reduction

in system dimensionality as the homogeneous analysis but with a much less

drastic assumption.

Measurement difficulties that arise in model verification of parameter esti-

mation using a heterogeneous analysis are also often cited as reasons for a

homogeneous analysis. Although theoretically it is possible to measure

separately the gas and solid temperatures, in practice it is difficult without seri-

ously affecting flow patterns within the reactor bed. These difficulties can be

minimized by using an internal thermal well as is common in industrial systems

and including this thermal well in the model development. This eliminates the

concern over whether temperature measurements within the reactor bed are

actually measuring gas or solid temperatures or a combination of both.

Another procedure often used with less severe assumptions than the equal-

ity of the gas and solid temperature is the pseudo-homogeneous analysis pro-

posed byVortmeyer and Schaefer (1974). This procedure has proven to be quite

effective for simple adiabatic packed bed analyses and involves reducing the

energy balances for the gas and catalyst to a single equation using the assump-

tion

(3.7-5)

Although this assumption is often not too restrictive, additional more severe

assumptions must· be introduced to reduce our mathematical model to the

pseudo-homogeneous form, since our reactor is nonadiabatic and requires

radial temperature considerations. A literature search failed to find any

 -326-

attempts of the Vortmeyer procedure for a nonadiabatic packed bed reactor.

For the full mathematical model developed in Section 3.3, the following assump-

tions would be necessary in addition to the standard Vortmeyer assumption

above:

1) Negligible energy accumulation in the gas phase.

2) No axial diffusion in the gas phase or

Lll)2Tg 1= o
az6z2]

3)

L 11 BT,l = LBr~
6rr6r
11 6Tg 1

4)

Obviously, these assumptions are too restrictive for standard nonadiabatic

packed bed analyses.

3. 7. 2. 2 Quasi Steady State Approximation

Transient analyses using the full. nonlinear model show that concentration

profiles reach a quasi steady state quite rapidly (often within 3 to 5 seconds);

whereas, the thermal response of the reactor bed is much slower' due to the

large heat capacity of the reactor bed and thermal well. An example of this
phenomenon is shown in Figure 3.7-7, where the transient responses of the solid

3. Thermal time constants are about two orders of In88nitude greater than the concentration time
constants.
 -327-

temperatures, thermal well temperatures, and concentrations are shown for a

major step change in the inlet gas temperature and inlet CO concentration. In

this example, the effect of the step change is nearly immediate on the concen-

tration profiles with the major effect being within the first ten seconds. How-

ever. Figure 3. 7-7a shows that the thermal well temperatures and the catalyst
temperatures take up to ten times as long as the concentrations to approach
the new steady state after the input step change. Note that the catalyst tern-

peratures shown in the figure are at the radial collocation point r =rc and that
the response of the catalyst temperatures near the center of the reactor (not

shown) is very similar to the thermal well response.

Furthermore, comparison of the thermal (vT) and the concentration (vc)

wave velocities as defined by Gould ( 1969)

p1 cp u 8 e
VT= a
p8 Cp1 (1-t)
(3.7-6)
vc = u,

verify this quasi steady state for the concentration profiles. As in many solid
catalyzed gas reactions. the ratio of the concentration wave velocity to the ther-

mal wave velocity is quite large:' This implies that the concentration profiles

reach a quasi steady state rapidly and that this quasi steady state then follows
the slowly chang~ng temperature profile, thus providing theoretical backing for

the transient behavior of the concentrations and temperatures in Figure 3.7-7.

Packed bed studies by Jutan et al. (1977) have also shown that the concentra-

tion dynamics can be ignored in the packed bed modeling.

The quasi steady state approximation then allows the concentration time

derivatives to be set equal to zero

4. About 250 for our reactor.

 -328-

ax~
- - R:l 0 (3.7-7)
at
in the original partial differential equations or in the linearized model, thus
reducing 2N of the discretized ordinary differential equations to algebraic equa-

tions. In the linear system, these algebraic equations can be solved directly for

the concentrations and substituted into the differential equations. Computer

programs (Appendix 4) were written to simulate the reactor using this assump-

tion for both the full and linearized models. Although this approximation intro-

duces considerable simplification in the nonlinear model, its effect on the linear-

ized system is a simple reduction ln system dimensionality, and it does not lead

to any major reductions in analytic solution times.

Simulations show negligible differences in the transient temperature and

concentrations profiles as a result of this approximation. Some of these simula-

tions are shown in Section 3. 7-4 where the model reduction procedures are com-

pared. The major advantage of this assumption should be apparent in control

system design where a reduction in the size of the state vector is computation-

ally beneficial or in the time-consuming simulations of the full nonlinear model.

It should be stressed that, although this quasi steady state approximation

involves setting the time derivatives of the concentrations equal to zero, it does

not imply that concentrations are independent of time, since concentrations are

still coupled directly to temperatures, and temperatures are obviously time

dependent.

3. 7.2.3 Negligible Energy Accumulation in Gas

Another potential model simplification procedure involves assuming negligi-

ble energy accumulation in the gas phase as compared to that in the solid. This
 -329-

· is equivalent to the approximation above (Equation 3.7-3) based on the relative

magnitude of the energy accumulation in the gas and solid. For our system, the

accumulation of energy in the solid is approximately 250 to 300 times that in

the gas phase due to the relative thermal capacitance of the gas (Equation 3. 7-

2) and the similarity of the temporal behavior of the gas and catalyst tempera-

tures (e.g., Figure 3.7-8). Thus the accumulation term in the energy balance for
the gas phase can be neglected:

(3.7-8)

in comparison to the energy accumulation in the catalyst and thermal well. This

reduces N of the original ordinary differential equations (after orthogonal collo-

cation) to algebraic equations. Again after linearization. these can be solved

directly for the gas temperatures which can then be substituted into the
remaining ordinary differential equations to eliminate the gas temperatures
from the state vector. Simulation computer programs were again written

(Appendix 4) and show negligible differences in the dynamic and steady state
profiles, small reductions in solution times for the nonlinear model. and no

major time reductions for the analytic solution. Some of these simulations are
shown in Section 3. 7-4.

3.6.1 llodel Reduction

Regardless of whether orthogonal collocation or orthogonal collocation on

finite elements is used for the discretization, the resulting linear state-space

representation is of high order (3D-40 states) 1 due to the original system of five
coupled partial dHierential equations and the accurate treatment of the gas.
catalyst and thermal well temperatures and concentrations. Although dynamic
1. This is actually of very low-order in comparison to traditional finite difference solutions.
 -330-

simulations using the state-space representation are very fast, computational

difficulties still exist in developing multivariable feedback control algorithms

due to the relatively high order of the system. For such applications, an accu-

rate reduced-order model is desired.

Significant reduction is possible through the introduction of the approxi-

mations of quasi steady state for the concentrations, negligible accumulation of

energy in the gas phase, or pseudo-homogeneity of the system. Although these

approximations could have been made in the original modeling, doing so without

a careful analysis of their effects on the model behavior can be dangerous. The

complete study of the mathematical modeling of a packed bed reactor allows

careful investigation of the significance of these assumptions and the combined

effects of several simplifications. Additionally, an analysis or the eigenstructure

of the system shows that simple modal reduction techniques such as those

presented by BonVin (1980), Gould (1969), and Wilson et al. (1974) can lead to a

lo¥.·-order state-space model. This section considers explicit modal reduction

approaches to model reduction that result in an explicit reduced model formu-

lation without statistical analysis of input! output data.

Various model reduction approaches, or minor modifications to existing

approaches, were proposed during the late 1960's and early 1970's. The "'asic

strategy of these approaches is to retain certain modes of the high-order model

in the low-order model. Wilson et al. (1974) summarized these techniques and

showed that many of the published modal approaches are equivalent since they

produce identical reduced models. They further considered the design of

reduced-order control models by performing the model reduction on the high-

order model and then designing a low-order controller and by designing a high-

order controller and then reducing this to a low-order control law. Bonvin

(1980) also provides a comparison of the various modal techniques with respect
 -331-

to their steady state and dynamic accuracies as well as to the dependence of the

reduced models on the retained state variables.

According to Gould (1969), the central theme in modal reduction and con,

trol is

''that the transient behavior of a process is predominantly

determined by the modes associated with the smallest
eigenvalues. If it is possible to approximate a high-order
system by a lower-order system whose slow modes are the
same as those of the original system, then attention can be
focused on the attempt to alter the eigenvalues of the slow
modes so as to increase the speed of recovery of the process
from disturbances. It is essential to be aware of the fact
that various disturbances excite the modes differently so
that a scheme which is based on a lower-order model may
be inappropriate if a disturbance injects most of its 'energy'
in a fast mode which has been neglected."

This basic approach is really diVided into several distinct categories. Two of

these, Davison's method and Marshall's method, provide suitable model reduc-

lion for the state-space representation of the methanation reactor to a 12th-

order model. Comparisons of the models and discussion of additional model

reduction are presented in the next section.

3. 7. 3.1 Davison s Method.

The principle of this method, proposed originally by DaVison (1966), is to

neglect eigenvalues of the original system which are farthest from the origin

(the non-dominant modes) and retain only dominant eigenvalues and hence the

dominant time constants of the system. If we consider the solution of the

linearized model derived in Section 3.6,

(3.7-9)

and let
 -332-

A = SAS-1 (3.7-10)

where A is the diagonal matrix of eigenvalues

A = (3.7-11)

0 0

and Sis the corresponding matrix of eigenvectors, the dynamic behavior of the

system is governed by the term

A-l(eAdt- I)lfwtt) = SA-t (eAdt- I)s-tlfwtt)

l.ldt
e -1
0
At

=s s-1Ww{t) (3.7-12)

e A,_dt - 1
0
An

Obviously by neglecting the non-dominant eigenvalues, the dynamic behavior of

the approximate system will be similar to the original system, since the contri-

bution of the unretained modes will only be significant early in the dynamic

response.

Table 3.7·1 shows the system eigenvalues for our full 30th-order linear

state-space representation for Type II (Table 3.4-4) conditions. As shown, the

eigenvalues can be grouped into five distinct groups based on the real parts of

the eigenvalues:
 -333-

Group -(Real)
Rj 50

II 1-5

III 4-6

N 0.06- 0.08
v 0.01- 0.03

Further analysis shows that the fastest modes (Group I) correspond directly to

the gas temperatures at the interior collocation points and those of Groups II

and III correspond to the concentrations. It is evident that, of the 30 modes of

the full linear model (with N =6), 18 are very fast in comparison to the remain-
ing 12 (by two orders of magnitude or more). Thus direct modal reduction to a

12th-order model using Davison's method should provide good dynamic accu-

racy.

T,
Group Real Part Part
-53.49 ± 1.93
I -52.93 ± 6.31
-51.98 ± 10.46
-5.03 ± 5.76
II -2.12 ± 10.38
- 1.37 ± 10.35
-6.27 ± 1.89
III -5.66 ± 1.85
-4.34 ± 5.68
-0.08 ± 0.010
N -0.07 ± 0.035
-0.06 ± 0.074
-0.033 0.0000
v - 0.021 ± 0.0011
-0.018 ± 0.0009
-0.016 0.0000

Table 3.7-1
Eigenvalues of Full 30th-Order linear Model

However by simply neglecting the non-dominant modes of the system, the

 -334-

contribution of these modes is also absent at steady state, thus leading to possi-

ble (usually minor) steady state offset. Several identical modifications (Wilson et

al.. 1974) to Davison's original method have been proposed by Davison and Chi-
dambara (Chidambara and Davison, 1967abc; Davison, 1968) and by Fossard

(1970). In these methods, the states of the reduced model are artificially recon-

ditioned to ensure desired steady state behavior.

3. 7. 3. 2 MarshaUs Method

Marshall's model reduction technique (Marshall, 1966) differs from

Davison's in that the steady state characteristics of the original system are

retained in the reduced model. Since the response of any element of the state

vector associated with a large eigenvalue is much faster than that of elements

associated with the smaller eigenvalues (or larger time constants), the dynamics

of the non-dominant modes can simply be neglected. 6 This is equivalent to

approximating the response of the faster modes by an instantaneous step

change. If the fast modes in the methanation model are taken as the gas tem-
peratures and the CO and C02 concentrations, Marshall's procedure is a rigorous

mathematical reduction identical to the assumptions of quasi steady state for

the concentrations and negligible energy accumulation in the gas phase. An

important advantage of Marshall's method over Davison's method is that the

reduced-order model has the same steady state as the high-order model. since

the time derivatives are identically zero for all state variables at steady state.

However, the retained modes may no longer optimally represent the dynamic

behavior.

Table 3.7-1 shows that, for the methanation reactor model, the dynamic

6. I.e., their time derivatives are set equal to zero.

 -335.

response of the gas temperatures and CO and C02 concentrations should be

much faster (by two orders of magnitude) than the response of the catalyst and

thermal well temperatures. This prediction is verified in the dynamic responses

shown in Figures 3. 7-7 and 3. 7-6 and the previous analysis or the thermal and

concentration wave velocities.

Thus the state vector

i = 1, ... , N (3.7-13)

is partitioned into

(3.7-14)

The state-space representation

i: =Ax+ .... (3.7-15)

A11 A12 A1s A14 Aus •~

Ae 1 Ae2 Aes Ae4 Aeo ...2

where A= As1 Aa2 Ass Aa4 Aao and W= Ws (3.7-16)

A.:1 A.:2 A.:s Aw ~ w4
.l\s1 .l\s2 .1\ss .1\54 .1\55 .5
can then be partitioned into two sets of equations:

[Au Ats] + [A12 A14 Ate~] (3.7-17)

*•= Aal Aas x1 Ae2 Ae4 Aa5 :1.2 + [::],

Aet Aes .2
- *2= A.:l A.s :x:l + ""'
A.:2 ""'
Aw ~A,,] Xe + .4 ... (3.7-18)
~

Ae1 Aos Ae2 Ae4 Ao5 ...5

Then we can let ie =0 in light of the quasi steady state approximation and the
assumption of negligible energy accumulation in the gas or in light of the
 -336-

significant differences in the magnitudes of the eigenvalues for the x:1 and x::z

states and solve for x::z as a function of x1 . Substituting this result into Equation

(3.7-17) results in a reduced state-space model

(3.7-19)

where the state variables arc now the catalyst and thermal well temperatures at

the collocation points and the new matrices K and lJ' are simply related to the
original matrices:

(3.7-20)

(3.7-21)

Ae2 Ae4 Ae~l

where Ac- 1 = ~2 ~ ~~ (3.7-22)
Ao2 ~ Ao~

Of course, this technique does not actually eliminate the states

9~, y 1 ~, and y9 . Instead it retains their steady state effects and relates their

dynamic behaviour to the gas and thermal well temperatures. The reduced

model is then of order 2N, or of only 12th-order for N=6. Similarly, we could con-

sider each assumption independently. If we retain the state variables 9", the

model is of 3Nth-order.

3.7.4 Di.scuasion of Reduced llodel

The reduced-order model obtained using Marshall's method is an accurate

2Nth-order approximation to the original bNth-order model. Although the

 ~ 337-

resulting model is equivalent to simply making the common assumptions of

negligible energy accumulation in the gas and quasi steady state for the concen-

trations, we have provided a rigorous mathematical approach to these assump-

tions based on the eigenstructure analysis and have provided an accurate

means of evaluating their applicability. The steady state problems associated

with Davison's method are eliminated, and the potential dynamic disagreement

between the original and reduced models is minimal for the methanation reac-

tor as verified by simulations.

Simulations using this reduced model show a reduction in computation

time, along with storage space, without any significant loss in accuracy. Table

3.7-2 shows the simulation times for various simulations using various models

and solution techniques with N =6. All computer programs are documented in
Appendix 4. The program RDt MOD simulates the 2Nth-order model using ana-

lytic solution of the equations, and RD2MOD simulates an 3Nth-order model

where only the concentration dynamics have been neglected. Although the solu-

tion time advantages between the analytic solutions of the reduced models and
the full linear model seem to be minimal, these analyses were conducted with a

constant control and disturbance vector 1r over the periods of disturbance or

simulation. If these values change frequently as may be the case in practice, the

solution time savings for the reduced models will be increased.

Figures 3.7-9 and 3.7-10 show compariSons or the transient gas and solid

axial temperature profiles for a step input change using the full model and the

reduced models. The figures show negligible differences between the profiles at
tL-rnes as low as ten seconds. Concentration results (not shown) show even

smaller discrepancies between the profiles. Additional simulations are not

shown since all attempted simulations showed minimal ditlerences between the

solutions using the different linear models. Thus for the methanation system,
 -338-

Model
Simulation
NLNMOD I.JNMOD ANAMOD RD2MOD RD1MOD
To ... 593°K 00:24
Step Xc;o.,..0.07 2:54:24 12:54 00:36 00:22
Step r.... o93°K 3:14:41 18:04 00:34 00:22 00:20
Start-up 4:43:26 28:39 00:29 00:17 00:14
Disturbance T0 ... 602°K 4:58:42 23:10 00:31 00:20 00:17
(50 seconds) xco ... 0.072

Table 3.7-2
Comparitive Simulation Times o! Models

Marshall's model reduction provides an accurate 2Nth-order reduced state-

space representation of the original5Nth-order linear model.

The excellent dynamic agreement between the original and reduced models

(Figures 3.7-9 and 3.7-10) can be explained by the eigenstructure of the reduced

system. As shown in Figure 3.7-11, the eigenvalues of the reduced model are

nearly identical (within 17.) to the dominant eigenvalues of the original model.
Thus the dynamic behavior is nearly identical to that which would result from

modal reduction using Davison's method. The advantage then of Marshall's

modal reduction for the methanation reactor model is that some contribution

of the 'fast' modes is still retained in the algebraic equations that result from

the assumptions. These contributions lead to small deviations in the remaining

eigenvalues and eliminate steady state discrepancies without seriously affecting

the dynamic responses.

Since the 2Nth-order reduced model based on Marshall's reduction pro-

ccdure accurately simulates the performance of the full linear model for a large

range of input changes and disturbances, there is little or no incentive to

attempt other techniques. If however further model reduction is desired or

necessary for control studies, more powerful reduction techniques would be

needed since the eigenvalues of the 2Nth-order model are of similar magnitudes
 -339-

and simple elimination of the larger ones may lead to major errors.

One possible procedure for further reduction would be Litz 1s modal reduc-

lion described in detail by Bonvin (1980). Litz proposed that the contribution of

the non-dominant modes be taken as a linear combination of the dominant

modes rather than simply being neglected. The appropriate linear combination

is determined in order to minimize the error between the responses of the non-
dominant modes in the original and in the reduced-order models. Bonvin (1980)

further explains that the eigenvalues of the reduce model are identical to the

dominant eigenvalues of the original model, but the eigenvectors are given a new
optimal orientation. Bonvin uses this technique to reduce a 24th-order model

for a tubular autothermal reactor to a 5th-order reduced model. He concludes

that Litz's procedure is superior to all other modal approaches.

 -340-

675

:X:
t.:)
650
w
0

w 625
a:
=:J
f-
a:
a: 600
w
ll..
::1:
w
f-

ff)
575
a:
t.:)

550
o.o 0.2 0.1! o.s 0.8 1.0
AXIAL POSITION, Z/L

Figure 3.7-1
Steady State Axial Gas Temperature Profiles
Standard Type I (Table 3.4-5) Conditions
Varying Axial Discretization

- - N = 8, 10, lZ - - - N=6 ----- N=4

 -341-

:s:::: 650
(.!)
li.J
0

li.J
a:
::::>
1-
a:
a: 600
li.J
CL
::E:
li.J
1-

550~--~----._----~----~----~----~----~----~--------~~
o.o 0.2 Q,ij 0.6 0.8 1.0
AXIAL PDSITIDN, Z/L

Figure 3.7-2
Dynamic Axial Gas Temperature Profiles
Standard Type I (Table 3.4-5) Conditions
Varying Axial Discretization

-- N = B. 10. 12 -~--- N=4

 -342-

-0.02
I.I.J
:::J
_J
a:
>
z:
I.I.J
(.!)
......
I.I.J

1.1....
0
1-
a:
a:
a..
_J
a:
w
a:
-0.06

• - REAL EIGENVALUE
~ - COMPLEX EIGENVALUE

-o.oa~~~~--~----~--~--~--~--~--~--~
0 2 ~ 6 8 10
NO. OF COLLOCATION POINTS

Figure 3.7-3
Convergence Pattern of Dominant Eigenvalues
Standard Type I (Table 3.4-5) Conditions
 -343-

-0.02
L.U
=>
-l
a:
>
:z
L.U

-
<...:)

L.U
IJ... -0.01.!
0
1-
a:
a:
0...
_J
a:
w
a:
-0.06

• - REAL EIGENVALUE
~ - CDMPLEX EIGENVALUE
-o.os~~~~--~--~--~--~--~--~---L~
o 2 "' 6 e 10
NO. ~F COLL~CRTION POINTS

Figure 3.7-4
Convergence Pattern of Dominant Eigenvalues
Standard Type II (Table 3.4-5) Conditions
 -344-

650

lJJ
a:
:::l
f-
a:
a: 600
lJJ
0..
:::E
lJJ
1-

SSOL-~L-~--~--~----~--~--~----~--~~
0.0 0.2 o.~ o.6 o.e 1.0
AXIAL POSITION, Z/L

(a)

...
'70 'a
:.:
0.3 3.05
C'\1
0 0
u u
z
0 L-__:::~---------c~.L - - - - 1 z
D
1-
0.2 3.00
1-
u u
a: a:
a: a:
1.1.. 1.1..
lJJ 0. I 2.95 w
...J ...J
D 0
:::E :s::

O.QL-~L-~--~--~---L--~--~--~--~~2.90
0.0 0.2 0.~ 0.6 0.8 1.0
AXIAL POSITION, Z/L

(b)

Figure 3.7-5
Axial Steady State Protlles - Homogeneous Model
Standard Type I (Table 3.4-5) Conditions

a) Temperatures
b) CO and C0 2 Concentrations
 -345-

r • R0
625
r = r
c

IJ.J
a:
:;:l
1-
cr:
a:
IJ.J
0....
::1: 575
w
I-

0.2 0.~ 0.6 0.8 1.0

AXIAL POSITION, Z/L

(a)

.
i'o
.....
0.7
I. 55
'o
-):

C\J
0 0
w w
z z
0 0
1- 1-
w w
cr: cr:
a: a:
lJ.... lJ....
IJ.J LLJ
.....J .....J
0 0
::1: 0.3 ::1:

0.2L---~~~~--~--~--~--~--~--~~1.~0
o.o o.2 o.~ o.s o.8 t.o
AXIAL POSITION, Z/l

(b)

Figure 3.7-6
Axial Steady State Profiles - Homogeneous Model
Standard Type 11 (Table 3.4-5) Conditions

a) Temperatures
b) CO and C02 Concentrations
 -346-

z= 0.38L

t.:)
w
0
630

620
/
/
/...- -- z = 0.17L

w
a: 610 ,...,...- .,..,.,...-.----~ ----
:::>
1--
a: /
a:
w 600 /
a.. /
:I:
w
/
1-
590

20 ijQ 60 80 100
TIME (5ECJ

(a)

-:0 "''o
-
0
X
0.7

0.6 ~----------~2---
co
1. 60
:1:

C\J
0
u u
z ~ 1.55
0
z
0
1-
0.5 C0
u
a: ~--------- -2- - - 1. so 1-
u
a:
a: a:
1.1..
Q,ijv~~·~--------------------~co~----~ 1.1..
w
...J
0
:I: 0.3

0.2~~--~--~--~--~--~~~~--~--~l.ijO
o 20 .. a so eo 1oo
TIME tSECl

(b)
Figure 3.7-7
Transient Responses to Step Input Change
Standard Type II (Table 3.4-5) Conditions
(To from 573° K to 593° K. ~o from 0.06 to 0.07)

a) Temperature
~- Catalyst - - - - Thermal Well
b) CO and C02 Concentrations
- - z 0.171 - - - - z = =L
 -347-

61.10

630 z = 0.38L
:::r:::
<...:)
w 620
----
Cl z = 0.17L

w
a:
::l
I-
610 ---------
a:
a: 600
w
CL
::E
UJ
I- 590

580
0 10 20 30 uo so 60
TIME (SECl

Figure 3.7-B
Transient Temperature Responses to Step Input Change
Standard Type II (Table 3.4-5) Conditions
(T0 from 573° K to 593° K. x& from 0.06 to 0.07)

- - Catalyst - - - - Gas
 650

w 625
a:
:::;,
t-
a:
a: 600
w
a..
X:
w
t-
575

0.2 0.~ 0.6 0.8 1.0

RXIRL POSITION. Z/L

(a)

<.:::>
w 650
Cl

w
a:
:::;, 625
t-
a:
a:
w
a.. 600
X:
w
t-

-
Cl

. .J
0
en
575

0.2 Q.ij 0.6 0.8 1.0

RXIRL PCISITICIN, Z/L

(b)

Figure 3.7-9
Transient Axial Temperature Profiles - Start-Up Simulation
Standard Type I (Table 3.4-5) Conditions

a. Gas Temperatures b. Solid Temperatures

- - Full Linear Model
- - - Quasi Steady State for Concentration
- - - - Negligible Accumulation of Energy in Gas
 -349-

t z

600
w
a: 10 sec
:::::>
1-
CI:
a:
w
a... 575
::E
w
t-
(/")
a:
t.:l

550~~---L--~--~--~--~~---L--~--~
o.o 0.2 0.14 0.6 0.8 1.0
AXIAL POSITION, Z/L

(a)

t.:l
w
Cl

w SOD
a:
:::::>
t-
CI:
a:
1..1..1
a...
::E
w 575
t-
o
-I
D
(/")

0.2 o.l4 o.6 o.e 1.0

AXIAL POSITION, Z/L

(b)

Figure 3.7- 10
Transient Axial Temperature Profiles- Start-Up Simulation
Standard Type II (Table 3.4-5) Conditions

a. Gas Temperatures b. Solid Temperatures

- - Full Linear Model
- - - Quasi Steady Slate for Concentration
- - - - Negligible Accumulation of Energy in Gas
 -350-

o. 10
e - FULL MCIDEL
0.08
• - REDUCED MODEL
0.06 w
::J
...J
a:
>
O.Oll z
w
<.:>
0.02 w
., LL
0
• I •• 0.00 f-
a:
a:
0...
-0.02
>-
a:
a:
-0.01.1 z
(.:>
a:
-0.06 2:
.......

-0.08

-0.10
-0.09 -0.07 -0.05 -0.03 -0.01
REAL PART ~F tiGENVALUE

Figure 3.7-11
Dominant Eigenvalues of the Models
Standard Type II (Table 3.4-5) Conditions
 -351-

Chapter4

RESULTS AND DISCUSSION

OF FUTURE EXPERIMENTAL STUDIES

 -352-

4.1 CONCWSIONS or CURRENT ANALYSIS

This thesis provides the basis for a concerted theoretical and experimental

program in multivariable process control structure design for packed bed chem-
ical reactors by presenting an in-depth control analysis of a practical, multivari-

able, distributed parameter system-the heat conduction problem defined by the

simple diffusion equation-using both frequency-domain and time-domain anal-
yses and the formulation, numerical solution, and analysis of a detailed model

for packed bed reactors, along with reduction to a low-order state-space

representation suitable for on-line process control.

The first portion of this work presented in Chapter 2 centers on the control

aspects of a one-dimensional, two-input heat conduction system. In particular,

an analysis of a number of multivariable process control strategies, including
non-interacting control, optimal control, inverse Nyquist array, and characteris-

tic locus techniques, is carried out theoretically. The significant results of this
study are discussed in Section 2.8 and only the major conclusions are outlined

here.

The study of the heat conduction system allowed for a careful study of vari-
ous control design techniques. Of all of the ones considered, the characteristic
locus procedure seems to have the most potential for this particular distributed

parameter system. This scheme provides a systematic, computer-aided design

strategy in terms of high and low frequency compensation that leads to an

excellent, proportional-integral controller. The study also considers the relation

between measurement structure and control system design. The choice of
measurements and their locations sign.iftcantly affects the optimal control

design and the usefulness of the different design techniques. The importance of
'extra' measurements-those in excess of the number required in the feedback
 ~ 353-

loop-for improving the system response is also shown. The additional i.n!orma-

tion on the process contained in these measurements can be used in a variety of

ways, including reducing the interaction in the control system through the tech-

nique of inner-loop decoupling which is proposed in this work. Finally, the

analysis of the heat conduction problem clearly shows the importance of an

accurate process model and the necessity of model reduction to a low-order
state-space representation for control structure design and implementation.

Based on these preliminary results, the second portion of this study pro-

vides a detailed mathematical modeling analysis of packed bed catalytic reac-

tors that significantly extends previous studies in the detail of the model and in

the consideration of all aspects of the model development and reduction to a

state-space control representation. Explicit consideration of common

simplifications is then presented, and the validity of these assumptions is

assessed in light of their effects and overall benefits. The general view that

modeling simplifications are desired since they lead to a reduction in numerical

solution effort is contested, and it is shown that many simplifications are no

longer necessary with to day's advanced computational capabilities. A unified

approach to dynamic reactor modeling is developed and its importance in the

accurate description of dynamic and steady state reactor behavior, in the inves-

tigation of reactor start-up or the effects of process disturbances, and in the

development of an accurate reduced state-space model for the design of control

structures to stabilize the reactor under various disturbances or to provide

optimal system recovery from input changes is shown.

The basic results from this modeling study can be broken down into four

major classifications which are now discussed in detail.

 -354-

Dynamic Modeling

A detailed two-dimensional. two-phase dynamic mathematical

representation of packed bed chemical reactors is developed without any

major a priori simplifications. The model accurately accounts for multi-

ple chemical reactions, axial and radial diffusion of mass and energy,

property variations due to temperature, pressure, and mole changes, and

an axial thermal well. Detailed analysis shows the importance of the

heterogeneous analysis for highly exothermic reaction systems, espe-

cially for control applications due to significant temperature differences

between the gas and catalyst as a result of differences in the heat tlow

through each of the phases radially to the cooling jacket. Although a

homogeneous analysis may be satisfactory for adiabatic or steady state

analyses. the equivalent reduction in system dimensionality and thus

solution time is available through other somewhat milder assumptions.

Radial temperature gradients were found to be important, although

radial concentration gradients were shown to be quite small. Radial

energy diffusion must be retained in the model since this is the basic

mechanism for transfer of energy from the reactor bed to the outer cool-

ing jacket. Although radial and axial mass diffusion are of minor impor-
tance if the aspect ratios for the reactor are sufficient, their inclusion in

the detailed mathematical model does not increase its dimensionality.

Finally, axial thermal diffusion should not be neglected since this

assumption can lead to inaccuracies and since eliminating the axial ther-

mal diffusion terms may significantly increase numerical solution time or

may lead to instabilities in the solution procedure.

 - :355-

Numerical Solution

The orthogonal collocation technique of reducing the partial

differential equations to ordinary differential equations is the best avail-

able method except for conditions with sharp profiles or abrupt protlle

changes. In such cases, the orthogonal collocation on finite elements

procedure is recommended. This thesis provides the first analysis of a

packed bed reactor using orthogonal collocation on finite elements.

These techniques are far superior to finite difference procedures since

only a fraction of the number of grid points is necessary, leading to a

model of much smaller dimensionality. The number of collocation points

necessary for accurate representation of the profiles has a major effect

on the dimensionality or the resulting system of ordinary differential

equations and thus on the numerical solution time. Simulations and

eigenvalue analyses using the model developed in this work show that as
few as six axial collocation points may be satisfactory for packed bed

analyses. Further analysis shows that, if the order of the approximating

polynomial is increased beyond that necessary for accurate representa-

tion of the solution, oscillations of the polynomials between the colloca-

tion points can become quite large. Since these polynomials are used for
approximating derivatives in the partial differential equations, these

oscillations or 'rippling' can be very detrimental to the solution.

Far fewer collocation points are required for the radial than for the

axial profiles. The radial temperature profiles are accurately modeled by

a quadratic function. Thus with the boundary points at the thermal well
and the out-er cooling wall, only one interior radial collocat~on point is

· needed for the thermal profiles. However, using one radial collocation

point for the radial concentration profiles inherently assumes a uniform

 -356-

radial concentration profile due to the zero tl.ux boundary conditions.

Although assuming a fiat radial concentration profile is not identical to a

cubic or quartic representation, differences are extremely small and

definitely negligible in light of the increased system dimensionality of a

higher-order representation.

In addition to the comparatively small dimensionality of the result-

ing ordinary differential equation system, other important benefits of the

orthogonal collocation scheme are that the collocation points and trial
functions are selected automatically and optimally and that the solutions

arc derived in terms of their values at the collocation points and at these

points the solution is exact. One of the drawbacks of the orthogonal col-
location on finite elements technique is the non-trivial problem of

optimal selection of the number and position of the elements.

Control Model

An accurate state-space representation including all expected con-

trol and disturbance terms is obtained through analytic linearization of

the reaction rate expressions, algebraic equations, and ordinary

differential equations obtained from the original mathematical model of

the packed bed reactor. Simple adjustments to this representation allow

specific control considerations. Simulations of this linearized state-space

model show it to be very accurate even at conditions relatively far from

the steady state around which the linearization was performed. For sus-

tained step input changes, relinearization around the new steady state is
probably de~ired. With the accuracy of this linearized model and the

significant reduction in solution time, there is relatively little incentive

for retaining the nonlinear representation in most considerations.
 -357-

Two techniques were found to be effective in reducing the dimen-

sionality of the linear state-space representation for on-line control.

Both procedures provided accurate model reduction to a 2Nth-order

model. 1 Direct modal reduction using Davison's ( 1966) procedure of

retaining only the dominant modes is satisfactory for the packed bed

model but can lead to possible steady state offset. Marshall's ( 1966) tech-

nique retains the steady state characteristics by neglecting the dynamics

of the non-dominant modes. The 2Nth-order model developed with this

procedure retains the catalyst and thermal well dynamics and accurately

simulates the performance of the full linear model for a large range of

input changes and disturbances.

Packed Bed Reactor Behavior

The dynamic behavior of the packed bed reactor is dominated by the

catalyst and thermal well due to their large thermal capacitance relative

to the gas phase. The importance of the thermal well is minimal at

steady state in terms of conversion within the reactor. Its significant

effect is in the dynamics of the reactor. It does have a slight effect on the

steady state temperature profiles by conducting heat axially.

Finally, simulations showed that radial temperature gradients are

significant and that a 'hot spot' develops in the bed under conditions of
rapid conversion. Furthermore, classical problems with 'wrong-way'

behavior where increases in the inlet gas temperature produce a decrease

in the outlet gas temperature are predicted. Obviously this can lead to

significant control difficulties if the control design is based on the outlet

gas temperature.

1. Where N is the number of axial collocation points.

 -358-

Thus the major development effort in this work has been the careful

mathematical modeling of the experimental reactor system and detailed

analysis of modeling assumptions, numerical solution techniques, parameter

sensitivity, model linearization, and model reduction using kinetic expressions

based on literature studies and preliminary experimentation and heat transfer

values obtained from standard correlations or published results. Other parame-

ters are calculated direclly lrom the physical properties ol the gas and reactor

system. Using these parameters, significant contributions in the area of packed

bed reactor modeling and control model development have been possible, along

with insight into expected operational and optimization difficulties of the

laboratory experimental reactor system. The generality of the model developed

in this analysis allows its use for various packed bed chemical reactors including

those under adiabatic operation and those without an axial thermal well.

As mentioned previously, this study really provides a basis and a unified

approach to packed bed modeling. An outline of this approach is as follows:

• The initial model development for dynamic and steady state analysis of

packed bed reactors should include detailed consideration of all major

physical and chemical phenomena within the system based on general

mass and energy balances. With current numerical techniques, minimal a

priari simplifications are needed. Include all dispersion effects, parameter

variations due to temperature, pressure and mole changes, and aXial varia-

tions in fluid velocity. The model should be heterogeneous and include

both the radial and axial analysis. All potential reactions should be

included in the model. If a thermal well is used for temperature measure-

ments, its analysis should also be included.

• Nondimensionalize the equations with respect to the inlet steady state con-
 -359-

ditions and the characteristic time for the reactor.

• Apply radial orthogonal collocation with one interior collocation point.

Then apply axial collocation using Lagrangian polynomials of degree N. 2

• Use standard numerical techniques to simulate the behavior of this full

nonlinear system of ordinary ditierential equations. Investigate the

predicted behavior of the packed bed system. Check the applicability of

the assumptions of negligible axial and radial mass diffusion.

• Linearize the reaction rate expressions, the algebraic equations, and the

ordinary differential equations around the steady state. Replace all vari-

ables with deviation variables. This should result in a general state-space

representation. Compare simulations of the linear model to the original

nonlinear model.

• Consider the optimal model dimensionality by comparing simulations with

differing number of axial collocation points and by plotting the behavior of

the major eigenvalues as a function of the number of collocation points.

• Reduce the dimensionality of the model if necessary using Davison's ( 1966)

or Marshall's (1966) modal technique.

• Finally, estimate the model parameters from experiments and compare

the model simulations to empirical results. Incorporate nonuniform radial

velocity profiles and more detailed pressure relations into the model if

necessary.

In addition to the analyses of control model development, the current stu-

dies provide the mathematical tools necessary to develop and study control

structures for the reactor system in our laboratory. These studies are currently

2. 11 sharp profiles or abrupt chal18es in the profiles are expected, use orthogonal collocation on
finite elements instead.
 -360-

in progress. The remainder of this chapter briefly outlines some of the experi-

mental work in progress along with the applications for the models developed in

this work.
 -361-

4.2 PARAIIETER ESTIIIATION AND IIODEL VERIJlCATION

Before control studies can be performed using the control model developed

in tbis work and before the implementation of these feedback control struc-

tures on the experimental reactor system, careful parameter estimation is

necessary. Although many parameters necessary for the mathematical model

can be calculated directly from physical considerations, the reaction and heat

transfer parameters must be measured directly for the experimental system. If

we neglect the radial and axial mass dispersion and the gas/solid heat transfer

in the axial boundary conditions and assume similar heat transfer coefficients

between the gas and solid and the reactor wall at both sides of the annulus (r ""

Ro and r =R1), the resulting dimensionless parameters in the model are shown

in Table 4.2-1. We can then divide the necessary input parameters for the model

into groups as shown in Table 4.2-2. Of course in addition to these parameters,

we do need to specify the operating conditions such as inlet concentrations,

pressures, cooling t'luid temperature and inlet gas temperalure.

The physical reactor parameters are measured for the experimental reac-

tion system. The void fraction must be empirically determined for the specific

catalyst used in the experiments due to differences in catalyst crushing. The

catalyst for the preliminary experiments is standard Girdler G-65 methanation

catalyst, and the thermal well is stainless steel (Type 304). Their physical prop-

erties are readily available as discussed in Section 3.4.1. Data are also available

tor the methanation and steam-shift reactions for heats of reaction and equili-

brium constants. As discussed in Section 3.4.1. linear regression of the data is

performed over the expected temperature ranges of operation to determine

linear temperature relationships for the heats of reaction and relationships

based on van't Hofl''s equation for the equilibrium constants.

 M362M

Aspect Ratios

axial n = 1.1 dp radial m = R1/ dp overall a =LIR 1

Axial Dispersion

Radial Dispersion

a
f3s =

Heat Transfer

U118L UuL
/'s = l't.s =
u&oYbP.cp. ( 1-t) VtPtep.,~
U!!IL U!IL
/'g = 'Yta =
u&ov'ti>~oC-pl v tPtcPt.u"'

Heats of Reaction

L( -AHs 2)~0axfl2ko8
uiocPa To

Reaction Coetncients

tpioUio
M&olp8 ( 1-t)xil2~k o6
a2 =
Bot Numbers

Table 4.2-1
Dimensionless Parameters
 • 663-

Reactor t L Rc Rt
Catalyst Cp. p. k. dp
Thermal Well CPt Pt kt

Heat Transfer u. UWB U.q Pers Pezs

~ x... ~ ~ ~=
Heat of Reaction AH:y1 AH:u:2 AHs 1 AHs 2
Equilibrium Constant ~1 ~I! l<Pst Kp!U!
Reaction Kinetics ko.u: Eall K1 K2
kos Eas r.
----•ruru-""'"
r2
Table 4.2-2
Input Parameters for the Mathematical Model

The remaining parameters for the mathematical description of the experi-

mental system are the heat transfer variables and the reaction kinetics. These

remaining parameters need to be estimated from preliminary experiments for

the specific catalyst and reactor bed. The strategy for this parameter estima-

tion is shown in Figure 4.2-1. Due to the large number of parameters necessary

for the mathematical description of the experimental system, simultaneous esti-

mation of all parameters from preliminary experiments in the methanation

.
reactor is not feasible. Instead the estimation is split into three stages. The
 -364-

kinetic parameters (and the actual kinetic expression) are first estimated using

a specially constructed kinetics reactor. This reactor system and the determi-

nation of the kinetic expression are discussed by Strand (1984). Preliminary

results based on limited experiments were discussed earlier, along with their

effect on the numerical solution of the mathematical model. The rate was found

to be very fast and highly temperature dependent, leading to potential

difficulties in numerical simulations due to very steep axial temperature and

concentration profiles and complete conversion early in the reaetor bed. These

difficulties can be reduced by using the orthogonal collocation on finite ele-

ments solution technique rather than simple orthogonal collocation. Further

experimentation is needed for accurate determination of the kinetics expres-

sions, especially in terms of reaction reversibility, temperature dependence, and

the importance of the steam-shift reaction.

The heat transfer parameters can first be estimated using experiments

without reaction in the bed. These can be performed by measuring temperature

responses in the thermal well, in the exit gas, and in the cooling fluid to input

(inlet gas temperature, cooling fluid temperature, or gas velocity) disturbances

or step changes.

Finally, all estimated parameters, except those that are found to be highly

insensitive, can be updated from experiments in the methanation reactor

operating under desired conditions. This parameter estimation is currently in

progress.

A sensitivity analysis was also performed to evaluate how accurately some

of the parameters must be determined. This analysis should be used in conjunc-

tion with a correlation analysis of the model parameters. The major heat

transfer parameters were increased and decreased by twenty percent individu-

 -365-

ally, and the kinetic parameters were changed by ten percent. The dynamic sen-

sitivity is evaluated by the etiect of the changes on the major eigenvalues;

whereas, the steady state sensitivity is evaluated by the effect of the changes on

the steady state profUes and the outlet temperatures and concentrations.

Table 4.2-3 shows results tor the standard Type II (Table 3.4-5) conditions.

The major parameters atiecting the dynamic behavior of the system are those

related to the thermal well, thus verifying earlier conclusions as to the impor-

tance of the well on the dynamics of the system. For steady state behavior, the

methanation reaction parameters are of most importance.

DYNAIIlC STF..ADY STATE

-Real >..1 -Real~ T.__ Xco___

+20% -20% +20% -20% +20~ -20% +20% -20%
u. 0.0165 0.0165 0.0172 0.0172 609 609 0.0370 0.0368
u... 0.0167 0.0164 0.0173 0.0170 609 609 0.0369 0.0369
u~ 0.0133 0.0141 609. 609. 0.0369 0.0369
Ap. 0.0165 0.0165 0.0172 0.0172 609. 609. 0.0369 0.0369
A.s 0.0165 0.0165 0.0172 0.0172 609. 609. 0.0369 0.0369
Au. 0.0164 0.0167 0.0170 0.0174 609. 610. 0.0369 0.0368
Atrg 0.0155 0.0177 0.0162 0.0163 605. 615. 0.0373 0.0364
Pens 0.0166 0.0165 0.0172 0.0172 609. 609. 0.0369 0.0369
P~r 0.0167 0.0164 0.0173 0.0170 616. 602. 0.0363 0.0376

+10% -10% +10~ ·10% +10~ -10% +10~ -10~

j:O][ 0.0160 0.0165 0.0172 0.0172 609. 609. 0.0366 0.0369
kos 0.0165 0.0165 0.0172 0.0172 612. 606. 0.0347 0.0392
Kt 0.0165 0.0165 0.0172 0.0172 609. 610. 0.0375 0.0362
~ 0.0165 0.0165 0.0172 606. 611. 0.0381 0.0355
Ea~~ 0.0166 0.0165 0.0173 0.0171 595. 622. 0.0473 0.0223
Eas 0.0165 0.0165 0.0172 0.0172 609. 609. 0.0371 0.0358

Table 4.2-3
Parameter Sensitivity
 -366-

Kinetics Reactor Methanation Reactor

(No Reaction)
ko .. • K I • Kz • EaM
u.CJ • u., • u.CJ • Pe,g. PeZCJ
kos • F I • F2 ' E as XSZCJ t Xgzs • x.,. • x.,CJ

Methanation Reactor

Update all parameters, except those

that are highly insensitive.

Figure 4.2-1
Parameter Estimation Strategy
 -367-

4.3 CO.NTROL IIODKL CONJlGURA.TION

In this section, we consider reconfiguring the general control model for a

specific multivariable control structure. The control configuration shown in Fig-

ure 4.3·1 is studied for a single pass configuration of the reactor system. Note

that we have also neglected the feed-effluent heat exchanger in this preliminary

analysis. Temperature measurements are taken at M points within the reactor

bed (using thermocouples within the thermal well). Let us consider the control

of the gas temperature at the reactor outlet and at an internal point (possibly
the 'hot spot') using the inlet process stream flowrate and the inlet gas tempera-

ture as the control variables and with expected disturbances in the inlet CO con- .

centration and the cooling wall temperature. In our experimental system, the

control of the process flowrate is really set point control of the three mass con-

trollers with the ratio of hydrogen to carbon monoxide being kept nearly con-

stant (at about 3 to 1). The control of the inlet gas temperature possibly

involves a PID scheme with the overall control network controlling the inlet gas

temperature setpoint and a local controller on the preheater controlling the

actual gas temperature.

Consider the reduced state-space model developed earlier

(4.3-1)

where

%1::: [9 11 , ... ,91tJi, 9t:t, ... ,9tNr

(4.3-2)
w::: [aw. v&o, 9 0, yf, yf, Yltao· YJi.· Y8H.r
and the A and W matrices are defined as before, Equations (3.7-20) and (3.7-21),

and all variables are deviation variables. The variables in the original control
and disturbance vector w then need to be related to the actual control or distur-
 -368-

bance variables. e.., and yf are disturbance variables, and }'it2 is directly related
to yf by yfl2 =?'Yf where 7 usually equals about 3. This is analogous to a process

where the COIBa stream comes from some other process upstream and the

ratio between the two remains nearly constant. The variables yl, yfl2o. and y&4

are not used as either control or disturbance variables in this analysis. The inlet

gas temperature 9 0 is an input variable or is related to the heat to the

preheater Qp. Finally, the gas velocity u#Io is directly related to the input tlowrate

and gas temperature:

(4.3-3)

where the variables in this equation are dimensional. After nondimensionaliza-

tion,

u = Fp (4.3-4)
lo P~o

or in terms of the input variables

(4.3-5)
\lao = 28-26-yX&yf

and af'ter linearization

(4.3-6)

where - Mao (4.3-7)

(J- (28-267X&) 2

Thus the original state-space representation is reconfigured to the form

i:=AI+Bu+Dd (4.3-8)

where u = [Fp Oar d =[e.., yfr (4.3-9)

 -369-

and all variables are nondimensional and are deviation variables. It the original

control and disturbance matrix W was defined as

(4.3-1 0)

then the B and D matrices are given by

aW2+Ws]
(4.3-11)
W4 +f'W7+fJ(26,X&)W2 J

Finally we need to consider the measurement structure. For measure-

ments at M points within the thermal well,

(4.3-12)

where ®1 is the dimensionless temperature of the thermal well at point z,-. Since

(4.3-13)

and

9t,O = 9o
(4.3-14)
etJ;+l
1
=-AN+I.N+l [ t ~+l.A:eu + ~+l,Oao]
Ji:=l

if we let the temperature of the thermal well at z =0 equal the inlet gas tern-
perature, the measurement relationship m terms of deviation variables LS

y=Cx+Fu (4.3-14)

where

c = [OwxN Lt]
(4.3-15)
n =[awx1 ~1
 -370-

AtJ+l,l ( ) AtJ+l,N ( )
ll ( Zt ) -A.. lN+l Zt lN ( Zt ) - A.. lN+l Zt
~+l.N+l ~+l,N+l

(4.3-16)

__, ~+1,0 ( )
lo ( ZMJ - A.. lN+l ZJl
~+l.N+l

Thus we have taken the general state-space representation described by

Equation (4.3-1) and reconfigured it to the appropriate control model for the
particular structure of interest. Consider the somewhat simpler control struc-

ture with only one manipulated variable, the inlet gas temperature, to simplify
the following discussion. The reduction of the general state-space representa-

tion to the specific control model is similar to the above analysis. If we then
consider one measurement at ~ =0.3, the B and C matrices for standard Type II
(Table 3.4-5) with N = 7 are
 • 371-

T
0.001 0
-Q.00008 0
0.00002 0
-D.OOOOl 0
0.000008 0
-o.000009 0
0.000009 0
B = 22.18 c = 0.018
-5.228 -Q.020
2.726 1.003
-1.733 0.014
1.254 -o.007
-o.970 0.005
0.903 -o.003

This example shows one of the potential drawbacks of the orthogonal collo-

cation procedure for reducing the partial ditl'erential equations to ordinary

ditrerential equations. The input matrix B multiplies the input by the appropri-

ate proportionality constants to indicate the direct etJect of an input change on

thE: state variables. From physical considerations, it is known that the only tern-

peratures within the reactor to be affected immediately by the inlet gas tern-

perature are those at the bed entrance. However as shown by the above B

matrix, all state variables (catalyst bed and thermal well temperatures) are

predicted by the model to be immediately affected as a consequence of the con-

tinuous nature of the polynomial approximations used along the length of the

bed. Of course, the major etfect of an input disturbance is still to the state vari-

abies closest to the bed entrance. Thus, instantaneous changes or sharp gra-

dients moving through the bed cannot be adequately represented by these poly-

nomial functions. Note that the etfect on the thermal well temperatures is

extremely high due to the assumption that the thermal well at the entrance of

the bed is in equilibrium with the inlet gas temperature. A simple change in this

boundary condition can reduce this effect 1! needed.

 ~ 372-

The C matrix further shows the relationship between the measurement at

("= 0.3 and the state variables. The coefficients in the C matrix show the relative

effect of each state variable on the measurement. As expected, the dominant

effect is or the thermal well temperature at the third collocation point (at

("= 0.297) since this is closest to the measurement location.

Although we have considered only two relatively simple examples, similar

analysis on proposed control structures will be necessary to develop the

appropriate control modeL This analysis further provides some insight into the

behavior of the experimental system and of the proposed control structure.

 -373-

Tl T2 •••
Reactor

Figure 4.3-1

Control Configuration
 -374-

As discussed earlier, this thesis provides a basis for a concerted theoretical

and experimental effort in multivariable process control structure design for

packed bed chemical reactors. Such a control study must center around the
basic elements of the control structure:

• the measured variables,

• the manipulated or control variables,

• the control configuratton connecting the measured and manipulated

variables, and

• the control logic determining the effects of the measurements on the

values of the manipulated variables.

In an experimental system such as our methanation reactor, these decisions are

by no means easy. It must be determined how to use the large number of poten-

tial measurements effectively. Are they all to be used? Or what subset is

optimum'? In our experimental system, the potential measurements include

temperatures at up to 24 axial positions within the thermal well. Temperature

measurements are also available elsewhere throughout the reactor system and

Within the cooling jacket and Dowtherm condenser. These latter measurements

probably are not too useful for the control structure but do provide additional

information on the operation of the process and can signal problems such as

lack of etiective flow through the cooling jacket. Another potentially important

control measurement is the effi.uent CO, C02 , and CR. concentrations. Currently

these are only available using a gas chromatograph With a delay of up to three

or four minutes. Since this is comparable to or slightly longer than the charac-

teristic time constants of the reactor, these concentration measurements may

be of little use in dynamic control but may be useful to update the model
 -375-

parameters periodically. Because effluent concentration measurements are

generally available in industrial processes and because they may significantly

affect control performance, it is planned that an on-line CO/C02 analyzer will be

added to the process.

The experimental system also has various possible control or manipulated

variables. These include the heat to the preheater, the fiowrate of the process

stream, the bypass around the preheater and feed-effluent heat exchanger, the

ratio of the H2 to CO in the inlet process stream. and the recycle ratio. The inlet

concentrations of other species and the cooling fiuid temperature could also be

manipulated if desired. Again a number of decisions must be made to select the

optimal subset of manipulated variables.

Even after these questions have been answered, careful consideration of the

control configuration and control logic is needed to design the control system to

handle the large number of potential disturbances (e.g., inlet CO concentration,

cooling fiuid temperature, fiowrates) and to meet the control objectives. For

our experimental system, these objectives may include maintaining product

quality, minimizing energy usage, minimizing hydrogen usage, and limiting tem-

perature excursions within the bed.

The first step in any proposed control research on a new experimental sys-

tem must be the application of existing control techniques to the system to

assess their performance. Once this has been atlempled, an important area of

concern that should be studied is that of measurement structures in industrial

systems. With all of the possible measured variables, decisions must be made as

to the optimal measurement structure. However, since the measurements are

such an important part of the overall control system, the measurement struc-

ture cannot be chosen independently but must depend on the choice or manipu·
 -376-

lated variables, the control configuration, the control logic, and even the control

objectives. Although various studies have been published on the measurement

problem in an open-loop framework (Aidarous et al., 1975; Amouroux et aL,

1976; Kumar and Seinfeld, 197Bab), the measurement location problem has

been virtually ignored with respect to closed-loop behavior. Some of the work

presented in Chapter 2 of this thesis provides a basis for research on the

interaction between measurements and control structure. This analysis should

be extended to the packed bed reactor with a detailed investigation of the

approach to optimal measurement selection in a closed-loop control

configuration.

Another potential control technique that should be considered is the evolu-

tionary control structure. This should not be confused with the self-tuning

regulator where the controller parameters are recomputed based on updated

values of the model parameters. What is proposed here is an extension to the

work by Alvarez et al. (1981). They investigated the feasibility of a variable

measurement structure for a tubular reactor in which the best set of tempera-
ture sensors along the length of the reactor is selected in response to changes

in the operating conditions. Moreover, the actual structure of the control sys-

tem may need to be altered in response to changes in the characteristics of the

process during operation. The packed bed reactor system developed by Strand

(1984) and modeled in this thesis is an excellent system for the study of such

evolutionary structures due to the large number of available measurements and

controls, the fully automated nature of the system, and the substantial variance

in the operating profiles. The mathematical model developed in this thesis

should allow considerable study of the evolutionary control problem prior to

actual experimantal implementation.

 -377-

Finally, the control of the packed bed reactor operating with energy and
product recycle should be studied. Such systems are common in industry but

have generally been neglected in dynamic and control analyses.

 -378-

APPENDICES
 -379-

APPENDIXl

COilPUTER PROGRAMS FOR HEAT CONDUCTION PROBLEM

Many computer programs were written for the analysis of the heat conduc-

tion problem for determining root-loci, inverse Nyquist plots, and characteristic

loci, for designing control strategies, for producing the necessary computer

graphics, and for simulating the behavior of the heat conduction system. Since

these programs are not unique but rather are based directly on published

theoretical works by various authors, the computer programs are not all dupli-

cated in this thesis. Rather they remain on the Caltech computer in the direc-

tory [RRK.HEAT].

However, two programs are presented in this appendix due to their impor-

tance in the characteristic locus control design for the heat conduction system.

The first program performs the frame alignment for the design of the high fre-

quency compensator. At high frequencies, it is desirable to reduce the misalign-

ment angles between the compensated system's characteristic direction set and

the standard basis vectors. This can be accomplished by designing a real com-

pensator KH that approximates the complex plane of Gp- 1(jc.>) at some high fre-

quency "'h· Program AUH, listed in Table Al-l, was written for this purpose and

uses the routine AlJGN to perform the actual alignment.

At low frequencies, the encirclement criterion should be satisfied and the

moduli of the characteristic loci should be large. This can be accomplished by

manipulating the loci with an appropriate approximately commutative con-

troller Kr.. where Kr. =M'k(s)B with A and B being the real frame matrices that

approximate the complex frame of the eigenvector matrix and dual eigenvector
matrix, respectively. Again the routine Al.JGN is incorporated into a program,
 -380-

AI.JL. for tbis purpose. This program is listed in Table Al-2. The elements of the
diagonal matrix rk(s) are then chosen on the assumption that Ch =~k1 , where QJ
and ~ are the eigenvalues ot the compensated and uncompensated system and
k1 are the elements of rk(s). The compensator JCL is thus used to insure stability
and integrity and to adjust the gains at low frequency to reduce interaction.
 -381-

C ~ERFORMS THE FRAME ALIGNMENT FOR DESIGN OF THE HIGH FREQUENCY COMPENSATOR
COMPLEX S,DEL,DET,BISg,z>,OISg,z)
DIMENSION YIZ,ZI,ZIZ,Zl,EVIIZl,EVRIZI,INDIZI,CIZ,ZI,OIZ,ZI
DATA PI /3.1.159265•/
IJ•l/IBJI.
READIS,3lXXI,XXZ,W
SO•SORTIZ. I
C SET UP QISI
AA•SQ*COSIPI*XXII
BB•SQ*COSIZ.*PI*XXII
CC•SQ*COSIPI*XXZI
DD•SQ*COSI2.*PI*XX21
S•CMPLXIJI.JI,IJI
CALL QSETIS,Q,AA,BB,CC,DOI
CALL IDENTIB,SJI,Z,I.SI
CALL ALIGNIQ,Y,ZI
PRINT I, W,YII,Il,Yil,Zl,YIZ,ll,YIZ,Zl
DO ISS 1•1,4
CALL ZEROIIEVR,ZI
CALl ZER011EVI,ZI
CALL ZEROIIINO,ZI
CALL ZER021C,Z,ZI
CALl ZER0210,2,21
Y1•1.
vz•t.
IFII .EQ. 2 .OR. I .EQ. 41Yl•-Yl
IFII .EQ. 3 .OR. I .EO . .tlVZ•-VZ
211,11•YI*((J,+SQ*AA+SQ*881*Yil,ll + 11.-SQ*AA+SQ*BBI*YIZ,IIl
ZI1,21•Y2*111.+SQ*AA+SQ*BBI*YI1,21 + 11.-SQ*AA+SQ*BBI*YIZ,ZII
ZIZ,li•YI*II1.+SQ*CC+SO*ODI*YII,II + 11.-SQ*CC+SQ*ODI*YI2.111
Zl2,2l•Y2*111.+SQ*CC+SO*DDI*YII,21 + 11.-SQ*CC+SQ*DOl*YIZ,ZII
CALL EIGENPI2,2,Z,EVR,EVI,C,O,INDI
PRINTZ, VI,YZ,EVRIIl,EVIIIl,EVRIZI,EVIIZl
1gg CONTINUE
STOP
FORMATI1HI,//////5X,'FREQUENCY •',F1 •. 4,/////ZJIX,'MATRIX KH',
• 5X,EIJI.3,4X,E16.3/34X,EIJI.3,.X,E1JI.3,///////I
2 FORMATI1.8'X,'Y1 .•',F5.1,3X,"YZ •',F5.1,6X,'ElliiENVALUES OF CBKH',
• 6X,E9.2,3X,E9.Z/6ZX,E9.2,3X,E9.2/////I
3 FORMATI2X,3FS.II
END
SUBROUTINE AllliiNIV,V,MI
C THIS ROUTINE ALIGNS THE REAL AND COMPLEX EIGENFRAMES
COMPLEX VISS,Zl
DIMENSION YIZ,Zl,AIZ,2l,BI2,21,CI2,21,012,2l,SI2,2l,EVRI2l,
EVI121,1NDI21
DATA EPS /I .• E-16/
DO I JIS I• I , 2
DO ISS J•1,2
IFIM .EO. 21 All,Jl•REALIVII,Jil
IFIM .EQ. Zl BII,Jl•AlMA~IVIl,~ll
IFIM .Ea. ll ACI,Jl•REALIVIJ,lll
1•• IFIM .EO. II ICI,JI•AIMAGIYIJ,Ill
CONTINUE

Table Al-l

Computer program ALIH

 -382-

DO
l•3-l
t••
I•l,Z
CALL ZEROIIEVR,Zl
CALL ZEROIIEV1,2l
CALL ZEROll IND,Zl
00 IBI J•I,Z ,
DO IBI 1(•1,2
C!J,Kl•AII,Jl*AII,Kl + Bll,Jl*Bil,Kl
DIJ,Kl•AIL,Jl*Ail,Kl + BIL,Jl*BIL,Kl
t•t CONTINUE
DET•DI1,1l*OI2,2l - D!l,2l*DI2,ll
lfiABSIDETl .LT. EPSl~O TO SBB
Sll,ll•IC!I,1l*OI2,2l - CIZ,Il*Dil,Zll/DET
Sll,2l•ICI1,2l*OI2,2l - Cl2,2l*Dil,2ll/DET
Sl2,1l•ICI2,1l*DII,ll - Cll,ll*OIZ,lll/DET
SIZ,Zl•ICIZ,Z>•D!l,ll - Cll,Z>•OIZ,lll/DET
CALL ZER021C,2,2l
CALL ZER0210,2,2l
CALL EIGENPI2,2,S,EVR,EVI,C,O,INDl
IF I EVItt 1 l .~E. EVR I 2 l lMM•l
IFIEVR12l ,,T. EVRilllMM•2
IFIM .EQ. 21 Vll,ll•Cti,MHl
IFIH .EQ. Zl VI2,Il•CI2,HHl
IFIH .EO. ll VII,ll•Cti,MHl
lfiH ,EQ. ll VII,Zl•CIZ,HHl
t•4 CONTINUE
RETURN
,.. PRINT lB. DET
RETURN
t• FORMATIIH1,3X,'HATRIX IS SINGULAR',SX,E1 •• 4l
END
SUBROUTINE ZEROIIA,Nl
DIMENSIOI'f AINl
00 IBB I•I.H
t•• AII>•B.B
ltETURN
END
SUBROUTINE ZEROZIA,N,Ml
OIHEI'ISION AIN,Hl
DO 111B I • 1 , H
DO IJJB J•I,M
All,Jl·B.B
1.. CONTINUE
RETURN
END

SUBROUTINE QSET!S,O,A,B,C,Dl
COMPLEX QISB,Zl,AIISB,Zl,BIIS.,Zl,Rkl2,2l
COMPLEX S,DET
DATA PI /3.141592654/
CALL IDENTIB1,5B.2,l.Bl
Qll,ll•l.IS + SQRTIZ.l*A/IS+PI**Zl +SQRTI2.l*B/IS+4.•,t••z>
Qll,Zl•l./S- SORTI2.l*A/IS+Pl"*2l +SQRTIZ.>*B/IS+4.*Pl**ZJ
QIZ,ll•l./S + SORTIZ.l"C/IS+PI•*Zl +SQRTIZ.l*O/IS+ •. *Pl**ZJ
012,21•1./S- SQRTI2.l*C/IS+PI**2l +SQRT!2.J•O/IS+ •. *PI••2l
IIETURH
EHO

SUBROUTINE lOENTIA,N,M,Zl
COMPLEX AIN,HJ
At I ,Il•CMPLXIZ,B.BJ
All,Zl•CMPLK! •••••.• ,
A(2,1l•CMPLXt•·····•>
AIZ,Zl•CMPLX<Z •••• ,
IIETURH
END

SUBROUTINE MATHULTIA,B,Nl,Ml,NZ,MZ>
COMPLEX A!Nl,Mll,BIN2,H2l,C1
Cl•AI 1,1 l
Atl,li•Cl*B(I,ll + All,2J•B!2,ll
Atl,Zl•Ci•B<I,Zl + All,ZJ•BtZ,Zl
CI•AIZ,ll
At2,1>•Cl*lll,ll • AIZ,Z>•BI2,11
AC2,Zl•Cl*lll,Zl • A!Z,Zl*ll2,2l
RETURN
END

Table Al-1 Continued

 -383-

C PERFORMS THE FRAME ALIGNMENT FOR DESIGN OF THE LOW FREQ. COMPENSATOR
COMPLEX S,DEL,DET,XI2,2l,QI2,2l,BI56,2l,V12l,C12,2l,AIS6,21
COMMON XX1,XX2,RH11,RH1Z,RHZ1,RHZZ,RL11,RL12,RLZl,Rl22,ALPHA
DIMENSION YIZ,Zl
LOGICAL*4 WANTX
WANTX•. TRUE.
READIS,41 IS,XXl,XXZ,W,ALPHA,RHll,RHlZ,RHZ1,RHZZ,RLll,RLIZ,RLZ1,
RLZZ
PRINT S,W,ALPHA,RL11,RL12,RH11,RH12,RL21,RLZZ,RH21,RHZ2
CALL !DENT IC,2,2,1.61
C SET UP QISI
S•CMPLl<I •.• ,1Jl
CALL QSETIS,Ql
CALL IOENT 18,56,2,1.61
C CALCULATE THE EIGENVALUES AND EIGENVECTORS
CALL EIGENCIZ,Z,Q,C,V,WANTl<,Xl
Xl•CABSIV1111
XZ•CABSIVIZll
PRINT 1, Vlli,Xl,VIZI,X2,XI1,1l,XI1,21,XI2,1l,XI2,2l
IFI IS .EO. 1 >STOP
C NORMALIZE EIGENVECTORS
OEL•CSQRTIXIl,ll**Z + Xl2,11**21
OET•CSaRTIXI1,2l**2 + Xl2,21**21
AI 1,1 l•XI 1,1 1/0EL
All ,2l•X( 1,21/Dl':T
AIZ,l l•XIZ,1 !/DEL
AI2,21•XI2,2l/DET
C ALIGN THE REAL AND COMPLEX EIGENFRAMES
CALL ALIGN IA,V,ll
PRINT 2, Yll,ll,Yil,Zl,YI2,1l,YIZ,21
CALL CSLECD IA,2,B,2,DET,ILLI
CALL ALIGNIB,Y,Zl
PRINT 3, V11,1l,VI1,21,VI2,1l,VI2,2l
STOP
FORMATI//////lBX,'EIGENVALUESt',
* 2C/3.8'X,FB.2,' + ',FS.Z,' t C',F7.2,')'),///1P,
* 'EIGENVECTORSt',2(/3BX,2<13P,2CF8.3.' + ',F8.3,' I',
• 3X I I I I
2 FORMATI/1///ZBX.'MATRIX 8',5X.E1B.3.4X.E1B.3/33X,E1B.3,4X,E1 •• 31
3 FORMATI/1///ZIX,'MATRIX A',5X,E11.3,4X,E16.3/33X,E16.3,4X,E1 •• 3.
• /11/l/1//)
4 FORMATI12,3F5.1/9F8.3l
5 FORMAT11Hl,/////I,SX,'FREaUENCY •',F8.4///15X,'ALPHA •',FS.l,
* SX,'KL •',2F9.3,5X,'KH •',2F9.3/36X,2F9.3,9X,2F9.3>
END

SUBROUTINE ALIGNIV,V,M~

C THIS ROUTINE ALIGNS THE REAL AND COMPLEX EIGENFRAMES

COMPLEX Vl5.,2l
DIMENSION Vl2,2l,AI2,2l,I12,2!,CI2,Zl,DI2,2l,SI2,21,EVR!Zl,
EVII21 ,INDIZI
DATA EPS /1.1E-36/
DO 1•• 1•1,2
DO 16• J•t,2
IF!M .EQ, 21 Ati,Jl•REALIVII,Jll
IFIM .Ea. Zl ~(I,Jl•AIMAGIV!l,Jll
l·FIM .Ea. 11 A(I,JI•REALIVIJ,Ill
IFIM .EQ, ll 81l,~l•AtMAGIVI~,Ill
1.. CONTINUE
DO t•.t 1•1,2
L•3-l
CALL ZER011EVR,2l
CALL ZERO!IEVI,Zl
CALL ZEROliiND,Zl
DO t•t J•1,Z
DO 1•1 k•l,Z
CIJ,ICI•AII,Jl*AII,K) + III,Jl*ICI,Kl
DIJ,kl•AIL,JI•Ail,Kl + lll,Jl*lll,KI
1•1 CONTI HUE

Table Al-Z

Computer program AlJL

 -384-

DET•DI1,1l*DIZ,Zl - 011,Zl*DIZ,1l
IFIABSIDETl .LT. EPSl~O TO s••
S I 1, 1 l • I CI 1, 1 l *0 I Z, Z l - CI Z, 1 l *0 I I, Z)) /OET
SII,ZI•ICII,ZI*DIZ,ZI - CIZ,Zl*DII,Zll/DET
SIZ,ll•ICIZ,ll*Dil,ll - Cll,ll*DI2,lll/DET
SIZ,Zl•ICIZ,Zl*DI1,ll - CI1,Zl*DIZ,1ll/DET
CALL ZEROZIC,Z,Zl
CALL ZEROZI0,2,2l
CALL ElGENPCZ,Z,S,EVR,EVI,C,O,INDl
lFIEVRI1l .GE. EVRIZllMM•l
IFIEVRIZI .GT. EVR11llMM•2
IFIM .EO. 21 Vtl,ll•CII,MMl
IFIM .EO. Zl VCZ,Il•CCZ,MMl
IFIM .EO. II YCI,ll•CI1,MMl
IFIM .EO. 11 Vll.21•CI2,MMl
If' CONTINUE
RETURN
Sf. PRINT 1/1, DET
RETURN
1• FORMATC////////,3X,'MATRIX IS SINGULAR',SX,E1 •• 4)
END
SUBROUTINE ZEROliA,Nl
DIMENSION AINl
1•• DO 111/1 I•I,N
AIIl•ll.ll
RETURN
END
SUBROUTINE ZEROZIA,N,Ml
DIMENSION AIN,Ml
00 IIIII 1-I,N
DO IIIII J • 1 , M
AII,Jl•ll./1
tn CONTINUE
RETURN
END

SUBROUTINE OSETIS,Ol
COMPLEX 012,2l,Ali511,Zl,BIISII,Zl,RKIZ,Zl
COMPLEX S,OET
COMMON XXI.XX2.RH11.RH12,RH2t,RH22,RL1l,RllZ,RL21,Rl22,AlPHA
DATA PI /3.141592654/
CALL IDENTIB1,5/1,2,1./Il
SO•SQRTIZ. l
RKil,ll•ALPHA*RLll/S + RHil
RKII,ZI•ALPHA*RLIZ/S • RHIZ
RKIZ,Il•ALPHA*RLZI/S + RHZl
RKI2,2l•ALPHA*RL22/S + RH22
AA•SO*COSIPI*XXIl
BB•SO*COSIZ.*Pl*XXll
CC•SO*COSIPI*XXZI
DD•SO*COSI2.*Pl*XX2l
011,1>•1./S + SO*AA/CS+PI**2l +SO*BB/IS+4.*P1**Zl
011,21•1./S- SQ*AA/IS+Pl**2l +SO*BB/IS+4.*Pl**2l
012,11•1,/S + SQ*CC/IS+PI**Zl +SO*DD/IS+4.*Pl**Zl
012,21•1./S- SO*CC/IS+Pl**Zl +SO*DD/IS+4,*PI**2l
CALL MATMULTIQ,RK,2,Z,Z.21
RETURN
END

SUBROUTINE IDENTIA,N,M,Zl
COMPLEX AIN,HI
All,1l•CMPLXIZ,f.lll
AII,Zl•CMPLXI/1./I,f,/ll
AIZ,Il•CMPLXI/1.6,11./Il
Al2,2l•CMPLXIZ,f./l)
RETURN
END

SUBROUTINE MATMULTIA,B,Nl,Ml,N2,M2l
COMPLEX AINI,M1l,BIN2,MZl,Cl
CI•AII,II
All,ll•CI*BII,ll + All,2)*812,1l
Al1,2l•C1*111,2l + All,2l*BI2,2l
C1•AIZ,l l
Al2,1l•CI*BII,Il + Al2,2)*812,ll
AIZ,Zl•Cl*lll,Zl + ACZ,Zl*l12,2l
II.ETURN
END

Table Al-2 Continued

 -385-

APPENDIX2

NORilALIZED PACKED BED REACTOR MODEL

The normalization of the original mathematical representation of the

packed bed reactor is based on

tt = !_ a=.!To • !!L
Pg =-
L Pr.o
~
r=L
Rt Y• = =o ·-~
c;,.- "Cp.
xro
tUg, u .... :M,
"6= - - vg=~ Mg =-
L u.., M&o

..
Note that in the following normalized equations the () is dropped from
.....
p8 , cp • and M8 and that
11

• M.,~·
P, = ai
....
p('=l
-r= -...- - 1
p('=O

Total mass conservation (continuity):

 -386-

Energy balance for the gas:

r=1.0

<"=1.0

Note that the gas heat capacity, cp , gas density, pg. and gas velocity, Vg, are
1

functions of position and time due to their dependence on mole changes,

pressure, and temperature.

Energy balance for the catalyst:

Using a similar analysis to that for the energy balance of the gas and after

assurmng constant physical properties of the solid phase,

 -387-

ae.
a:;=-=>-trs(9s-Bt)
Bas
r=l.O -a:;=-=>-"n(e.-e")
aes
r-o "'B( =Aszg( 9s-6 g)
Be.
-ar=>-szg(es-a8 )

Energy balance for the thermal well:

Assuming constant physical properties in the thermal well,

(=0

(=:1.0

Mass balance in the reactor section;

where
 -388-

8y, = 0
Br

r=l.O Byi ::::0

Br
Byl VII( Yt 0)
~0 --:::: -Iii- -y~
0{ am
~1.0
By1 =0
B{

In these equations, the dimensionless quantities are defined as follows.

Aspect Ratios

axial n =L1 dP overall a = LIR1

Axial Dispersion

a. =

Dz
= --=
UIio

Radial Dispersion

a
 -389-

a
fJm = =

Heat Transfer

u.L UtJ.,
?'. :;:
UaoYbPsCp1 ( 1-e) ?'ts = vtPtCptilg.,
U!!IL U!iL
?'g = llgoYt/5g/jP/ ')'tg :;:
VtPt,cPt'Ugo

Heats of Reaction

L( -6Hs2)"X8o 2Xl'{2ko 5
~ =
ilg.,Cp1 T0

Reaction Coetlici.ents

~p&ou&o M&olp,( 1-e)xll2k 05

Mg04os( 1-e)xf1ex&2k 05 ep&oUP.o
0'2 =

Biot Numbers

Xgas =~
kzs
Xazg =~
kq

Awn =
hwsRl
kn
Awrg = hwgRl
krg
 -390-

Reaction Rates

5
(xH )· 15 x&kayR'y
Rw = p,( 1-e)P-t\~ 8
Rs = Ps( 1-e)X'&XHf:o5R's .
 -391-

APPEND1X3

RADIAL COU.OCATION OF PACKED BED REACTOR MODEL

Gas Phase Radial Temperature Profile

At the three radial collocation points r = ~ 0 • rc, and 1.0, let the gas tern-

peratures be a&o. a,!'. and a,l and assume that the radial profile is quadratic:

This profile must satisfy the boundary conditions

ae,
8r
l - Atrg(9
r-,o - ir..,o
-e)t
~ l
89 r-1 = -A,n.g(B&r-1-9,..)

where Atrg and Awrg are the dimensionless radial Biot numbers at the thermal

well and cooling wall, respectively. The profile must also be exact at the three

collocation points:

98o =do+ dl~O + d2~~

. e&r=da+dlrc+~r~
9g1 = do + dl + d2

After rearranging and eliminating 98o and 9 81 •

The expressions for di(~() in terms of air. at. and a,.. can be simply obtained by
applying Cramer's Rule. Let
 -392-

Then

do = w19gr + w29t + w39,

dl = w..,e&r + w 9t + w 9,
5 6

d:a = W79gr + Wa9t + Wg9w

where

Then based on the assumed quadratic profile

 -393-

the dimensionless form of the energy equation for the gas becomes

where 9 8 and 8 11 are now the temperatures at the radial collocation point rc, and

Pg and cp are dimensionless parameters, normalized with respect to the inlet

8

steady state values. Then if we let

G.>• = 4{1,w7 + f!.Lw..,

rc
- "Yg
~ =4(38wa + f!.Lwo
rc
c.>s =4f3i3,w + !!..La,w
9
rc 6

the energy equation for the gas becomes

Catalyst Phase Radial Temperature Collocation

Similar results are obtained for the energy balance of the catalyst using the

expression

The values ford;.(~~) in terms of 9~ar, 9t, and e, are derived as before. Let
 -394-

Then

do = w' 1 9~r + w' 29t + w' 39..,

d1 = w'.ell:-+ w'ts9t
+ w'e9w
~ = w'78fl:. + w'a9t + w' 99..,

where

w' _ At.rs(.\wrs+l) - Awn.(IPoA.u-s-1)

7 - det
w' _ At.rsXwrsrc - Xtr8 (Xwrs+l)
a- det
, Xwrs(IPoAtrs-1) - A.trsXwrrif'c
Wg = det
 -395-

The energy equation for the catalyst becomes

Thermal Well Energy Balance

We can now relate e.__

,....,.o
and e.."'Z""f'
. o to the gas and catalyst temperatures at

the radial collocation point. If we let

!.>7 ='lt.s(w'l +w'49'o +w'?9'8)

Wa ='lt.g(wl +w49'0 +w7rp6)
CJg = 'lts(w'2 + w'o9'o +w'a9'§ -1) + )'t,g(w2 + Wo9'o +wa9'~ -1)
Wto = 'lts9.,..(w's +w'a9'o +w'g,o~) + )'t.g9.-(ws +Wat;Oo + Wgt;O~)

the energy balance for the thermal well becomes

Radial Concentration Collocation

For radial concentration profiles, a quadratic representation may not be

adequate since application of the zero fiux boundary conditions at r = 9'o and

r =1.0 leads to d 1 -= d 2 =0. Thus a quadratic representation for the concentra-

tion profiles reduces to the assumption of uniform radial concentrations. which

for a highly exothermic system may be significantly inaccurate. Although

 -396-

additional radial collocation points greatly increase the dimensionality of the

resulting model, they may be necessary to accurately express the radial con-

centration profiles. A detailed analysis of multipoint radial concentration collo-

cation is presented in Section 3.4.5.

-397-
 -398-

APPENDIX4

DOCUM:ENTATION

COMPUTER MODEL OF THE PACKED BKD REACTOR

A large variety of programs have been written to model the methanation

packed bed reactor and to produce computer graphics of the various profiles on

a Zeta plotter or on the Versatec. This documentation describes these pro-

grams, the necessary library routines and input data and the command

sequences for operation of the programs. This description is complete only for

the latest revision (version 4). All previous versions may have some major

di.fferences. This revision has been extensively tested under most expected

operating conditions. The programs are located in directory [RRKMOD4). All

important modeling programs are also included in this appendix.

Although the routines involve excellent numerical solution techniques, they

may still experience numerical solution difficulties in some cases, due to the

extreme complexity of the reactor model. Various parameters have been

included so that the user can circumvent these numerical problems, but prac-

tice is necessary. In most cases, numerical difficulties will only occur during the

steady state solutions due in large part to very bad input profiles for the initial

guesses.

IIODELING ROUTINES

A two-phase, two-dimensional dynamic model was employed in this analysis,

with the assumption that the packed bed may be treated as a continuum insofar
 -399-

as changes occur smoothly and continuously Within each phase throughout the

bed. This assumption should be valid under the conditions of the methanation

reactor (Hlavacek, 1970) and allows for treating heat and mass fluxes in a form

analogous to Fourier and Fick laws.

The original modeling analysis includes all major expected phenomena in

the reactor bed. The model accounts for axial and radial dispersion of mass and

energy. for mole changes that occur along the bed due to the methanation reac-

tion, and for temperature. pressure, and mole dependencies of gas velocity. den-

sity, average molecular weight, and heat capacity, reaction rate constants, and

heats of reaction. Additionally, a central axial thermal well is completely

modeled, including axial conduction of energy along the well. Finally, the model

is based on a three-dimensional (time, axial, and radial) heterogeneous analysis

and incorporates the effects of axial pressure gradients.

The primary reaction in the analysis is the methanation reaction with the

steam-shift reaction being the only significant side reaction. These reactions

have been studied extensively and kinetic information is available. A rate

expression for the methanation reaction for conditions similar to those in the

present reactor is given by Lee(1973) and Vatcha (1976). A complete analysis of

the steam-shift reaction is provided by Moe (1962). Although these rate expres-

sions have been incorporated into the computer model of the system. the pro-

grams are written in such a way that the rate expressions can easily be changed

simply by replacing one subroutine.

The only major assumptions underlying the original model are:

• Reactor wall_ temperature is equal to the cooling tl.uid temperature and is

independent of length along the reactor.

 -400-

• Gas properties are functions of temperature, pressure and total moles as

dictated by the ideal gas law.

• There is no radial velocity.

• Global rate expressions are valid.

• Physical properties of the solid catalyst and thermal well are constant.

• Heats or reaction and gas heat capacities are described as linear func-

tions of temperature.

This original model was coded into the programs 110DEL.YOR and DIST.YOR

(found in directory [RRK.MODl] on the CHEMVAX). Further analysis using

dynamic simulations to study the effect of making several major simplifications

to the model structure are coded into the following programs (also in

[RRK.MODl]).

a. Negligible energy accumulation in the gas. (CilODELl, Clll)

b. Quasi Steady state for concentration. (Cil0DEI2, Cll2)

c. Homogeneous analysis. (CIIODEL3, Cll3)

It was concluded from these analyses that the first two assumptions may be

quite useful for later model reduction since they lead to minimal inaccuracies in

the simulated profiles.

Considerable ap.alysis using these complete modeling algorithms, led to the

following conclusions.

a. The temperature dependence of the heat capacity has very little effect on

the simulations.

b. Mass diffusion has little effect.

 • 401-

c. The central axial thermal well has little effect on the concentration

profiles but significantly alters the transient temperature responses.

d. Numerical stablity of the model solution is greatly enhanced by retaining

the thermal dispersive effects.

e. The assumption of pseudo homogeneity of the system leads to significant

discrepancies in the dynamic and steady state protlles.

Finally, a very detailed and careful analysis was made concerning the radial con-
centration profile. The programs for this analysis are found in [RRK.CONC]. All

of these programs use the full nonlinear model with no additional assumptions.
A lengthy analysis concluded that the assumption of constant radial concentra-

tion profiles in the reactor bed leads to inconsequential differences in Lhe axial

bulk concentrations, radial temperatures and axial temperatures.

Thus the final models described in this report include the additional

assumptions of constant heat capacity, negligible axial and radial mass diffusion

and constant radial concentration profiles.

LA Solution Strategy

The basic relationships taken to describe the system are the continuity
equation, the energy balance for the catalyst, the energy balance for the gas,

the energy balance for the thermal well, mass balance for CO, mass balance for

C02 and relationships for the density and pressure changes. These equations

are first normalized With respect to the steady state inlet conditions.

The resulting system consists of six coupled, three-dimensionaJ.. nonlinear

partial differential-equations, which must be solved to obtain the temperature

profiles in the gas, catalyst and thermal well, the concentration profiles and the

velocity profile. The technique of orthogonal collocation is used to reduce the

 -402-

equations to a set of first-order ordinary differential equations in the time

domain.

A one point radial collocation is performed. This is quite accurate for tem-

perature profiles since it assumes a quadratic radial representation of the

profile. Since the assumed quadratic profile must satisfy the boundary condi-

tions and since the concentration boundary conditions are zero flux conditions,

the use of one point radial collocation implicitly assumes constant radial con-

centration profiles.

Discretization of the resulting system is then performed by orthogonal col-

location in the axial direction. Since the position and number of points are the

only factors affecting the solution obtained by orthogonal collocation, any set of

linearly independent polynomials may be used as trail functions. The Lagran-

gian polynomials of degree N based on the collocation points z.t are used in this

analysis.

The resulting equations are a set of SN+ 1 ordinary differential equations

along with eleven algebraic boundary relations, where N is the number of axial

collocation points. The solution procedure is further simplified by solving the

continuity equation for the velocities as a set of algebraic equations, using tem-

perature values and temperature derivatives obtained from the solutions of the

remaining differential equations. Additionally, simple algebraic manipulation

allows for explicit solution of seven or the boundary variables. Thus the result-

ing dynamic model consists of a set of 5N coupled ordinary differential equa-

tions and N+5 coupled algebraic equations.

The solutions _of the model are obtained using modified Caltech library rou-

tines. The system of ordinary differential equations is solved using a

modification of the Caltech routine MODDEQ that uses an Adams-Moulton

 ~ 403-

predictor-corrector technique, with the method ot Runge-Kutta-Gill being used

to start the integration process. The nonlinear system of algebraic equations is

solved using modifications of Caltech's routines NSESJ and NSES2. The first of

these uses the standard techniques of inverting the Jacobian matrix but calcu-

lates the Jacobian numerically rather than having it input by the user. The

second uses Brown's (1967) quadratically convergent algorithm. NSESJ is a fas-

ter routine but is not as numerically powerful as NSES2. The modeling pro-

grams are written to allow the user to specify which of the two routines to use to

calculate the steady state solution. The dynamic solutions always use NSESJ.

LB Modeling Prugra.ma

Several programs were written to perform various types of modeling ana-

lyses. All of the programs have been extensively tested and are completely com-

patible. I.e., they use the same data tiles, calling formats and library routines

and have nearly identical output formats for easy comparison. All of the rou-

tines allow for completely arbitrary disturbances and step changes of the cool-

ing wall temperature, linear gas velocity, inlet gas temperature and inlet concen-

trations of the methane, carbon monoxide, carbon dioxide, water and hydrogen.

The routines are very modular, allowing for simple modifications and replace-

ments or reaction equations, output formats, input formats, modeling equations

and even solution strategies.

Difficulties may arise during some simulations due to very steep protiles or

disturbances. The collocation solution technique has problems with these situa-

tions since it assumes a smooth continuous protile along the system Sharp dis-

turbances or steep- temperature protiles in the system can lead to oscillations in

the calculated axial profiles. These problems can usually be reduced by increas-

ing the number of collocation points.

 -404-

The programs simulate the behavior of the model where the concentrations

are normalized with respect only to the inlet steady state concentration of CO.

This is better than normalizing with respect to the inlet steady state concentra-

tion of the indiVidual species in case one or more of the other inlet steady state

concentrations are zero.

a. NLNIIOD

This program simulates the complete nonlinear solution of the system. The

above techniques are combined with a variable time-step analysis to

efficiently obtain the dynamic and steady state reactor responses This vari-

able time-step procedure is automatic and involves increasing the time steps

as the system approaches steady state since the derivatives become smaller.

This program must be linked to MLIB and ILIB. A listing or this program is

included in Table A4-1.

b. LINMOD

This program performs the simulation for the linearized version of the

model, where the model is linearized about the steady state solution. The

program thus first calculates the initial steady state solution. based on a

user-defined initial guess. The program uses the standard numerical tech-

niques used in Nl.N110D. including the variable time-step analysis.

The linear system is of order 5N With the states being the solid tempera-

tures, gas temperatures. thermal well temperatures, CO concentrations and

C~ concentrations at the collocation points.

This program must be linked to III1B and UJB. A listing of this program is
 ~ 405-

included in Table A4-2.

c. ANAIIOD

This program performs the same simulation as LIN110D except that it uses

the analytical solution of the linear equations rather than performing a

numerical solution. Thus it takes only a fraction of the solution t1me and

does not need the variable time-step analysis. The solution of the equations

can be obtained explicitly at any desired time.

This program must be linked to 1lLIB, EIJB and LIJB. A listing of this pro-

gram is included in Table A4-3.

d. RDlllOD

This program performs the simulations for the reduced linear model using

the analytical solution. The assumptions o! negligible energy accumulation

in the gas and quasi steady state for the concentrations are used to reduce

the model to an order of 2N. The retained states are the solid temperatures

and thermal well temperatures at the collocation points.

This program must be linked to 11L1B, EilB and LIJB. A listing of this pro-

gram is included in Table A4-4.

e. RD2110D

This program performs the simulations for the reduced linear model using

the analytical s_olution. The assumption of quasi steady state for concentra-

tion is used to reduce the order of the model to 3N. The retained states are

the solid temperatures, gas temperatures and thermal well temperatures

 -406-

at the collocation points.

This program must be linked to llLIB, ELIB and um. A listing of this pro-

gram is included in Table A4-5.

LC Input Variables

The input variables are divided into groups and are described below. The

group names given between slashes are the names of the COMMON arrays that

the variables are stored in after they are read from the data file.

/REACP I - REACTOR PARAMETERS

EPS void fraction (cm.. 3 void/cm..3 bed)

L length of reactor bed (em)

RO radius of thermal well (em)

R: radius of reactor bed (em)

!CATlS! - CATALYST PARAMETERS

CPS heat capacity (callg 0 K)

PS density (g/cm••s catalyst)

TC thermal conductiVity (cal/sec em QK)

DC characteristic particle diameter (em)

/THWEL/ - THERMAL WELL PARAMETERS

CPT heat capacity (cal/g 0 K)

PT density (g/cm••s)

KT thermal conductiVlty (cal!sec em o:K)

 -407-

!GASPAI - GAS PARAMETERS

PTZ axial thermal Peclet number

PTR radial thermal Peclet number

UGS steady state inlet velocity (em/sec)

/HEATT I ~ HEAT TRANSFER PARAMETERS

OHSG heat transfer coefflcient(solid~gas)(cal!sec oK)

OHTS heat transfer coefflcient(wall-solid)(cal!sec 0 K)

OHTG heat transfer coefflcient(wall-gas)(callsec 0 K}

BSG Biot number (solid-gas)

BGS Biot number (gas-solid)

BTS Biot number (wall-solid)

BTG Biot number (wall~gas)

!OPCON I - OPERATING CONDITIONS

SCH4 steady state inlet methane mole fraction

SCO steady state inlet carbon monoxide mole fraction

SC02 steady state inlet carbon dioxide mole fraction

SH2 steady state inlet hydrogen mole fraction

SH20 steady state inlet steam mole fraction

PTO total inlet pressure (atm)

PTl total outlet pressure (atm)

STO steady state inlet gas temperature ("K)

 -408.

STW cooling wall temperature (aK)

/REAC 1I - REACTION PARAMETERS

DHlA heat of methanation constant, DHl=DHlA.-r+DHlB

DH1B heat of methanation constant (cal/mole)

DH2A heat of water-shift constant, DH2=DH2A•T+ DH2B

DH2B heat of water-shift constant (cal/mole)

KOP Arrhenius constant, methanation

KO Arrhenius constant, water-gas shift

IREAC2/ -ADDITIONAL REACTOR PARAMETERS

KP:A equil constant, methanation, ln KP1 =KP1A+KP1B/T

KPlB equil constant, methanation

KP2A equil constant, water-gas shift lnKP2=KP2A+KP2BIT

KP2A equil constant, water-gas shift

K2 constant for methanation reaction rate

K3 constant for methanation reaction rate

EAl activation energy, methanatton (cal/g mole)

EA2 activation energy, water-gas shift (cal/g mole)

OTHER INPUT PARAMETERS:

DTO time step (sec)

N number of axial collocation points

TMAX maximum time (length or simulation) (sec)

 -409-

RR radial collocation point with 0 < rr < 1

NEP precision in numerical routines, lO ..nep

IFLAG type of initial profile guess

0 values input for canst temps, cone, vel

1 entire profiles input

NFLAG steady state algebraic equation solver

1- NSES1, 2- NSES2

DL length of disturbance (sec)

IF max time step for variable time stepping is DT0•4••(IF-1)

NP number of time steps between printing

These variables are read in under the following format:

READ 1. EPS,L,RO,R:

READ 1. CPS,PS,TC.DC

READ 1, CPT,PT,KT

READ 1, PTZ,PTR,UGS

READ 1, OHSG,OHTS,OHTG,BGS,BSG,BTS,BTG

READ 1. SCH4,SCO,SC02,SH2,SH20,PTO,PT1,STO,STW

READ 1, DH1A,DH1B,DH2A.DH2B,KO,KOP

READ 1. KP1A,KP1B,KP2A,KP2B,K2,K3,EA1,EA2

READ 2, DTO,TMAX.RR,DL.N,NP.NEP,IFI.AG,IF.NFLA.G

where format 1 is (9!8.2) and format 2 is (4f8.2/6i8). All of these variables may

not be necessary for all of the programs but to keep consistency in the input

data tiles they are all included.

 - 410-

Note: Two of the variables change meanings in the different programs so that
the values in the data tile do not have to altered and so that the outputs

of the programs will be identical. These are DTO and NP. In the pro-

grams that require numerical solution (NLN110D, UN110D), DTO must be

taken very small (around 0.005 second) and then NP is set large so that

the profiles are only printed out for every five or ten seconds during the

simulation (i.e. NP= 1000). However, in the programs that are based on

analytical solutions, it is not necessary to take such small steps. If

printouts of the solutions are desired for every 5 seconds, then the time

steps should be set at 5 seconds. However, this would involve changing

the data files if you wanted to run say NLNliOD and then compare it to

ANAIIOD. To eliminate this problem, the actual time steps used in the

analytical programs are DTO•NP.

After this set of inputs, the user must specify the initial guesses for the steady
state calculation as per the type specified by IFLAG:

!FlAG= 0

Input the constant gas temperature, solid temperature, thermal well

temperature. CO concentration, C02 concentration and velocity under

format (6f8.2).

IFLAG = 1

Input the gas temperature, solid temperature, thermal well tempera-

ture, CO concentration, C02 concentration and velocity at each collo-

cation point under the format (6f8.2) and then the solid tempera-

ture at z=O and z= 1, the gas temperature at z=O and z= 1 and the

velocity at z=O under the format {5f8.2). The temperatures must be

 -411-

entered in nondimensional terms (as is output by the programs) and

the concentrations must be based on the inlet number of moles (as is

the output). Thus these inputs can be directly taken from one of the

output profiles.

Then !lnally, the type of simulation must be entered. The next input JFIAG tells

what type of simulation is being run. This variable is entered under the format
(i2).

JFIAG =0 (step or disturbance)

The step or disturbance vector is then input. The inlet gas velocity,

inlet gas temperature, CO concentration, C02 concentration, H2 0 con-

centration, H2 concentration, CJ-4 concentration and cooling wall tem-

perature are input under the format (8f8.2).

JFLAG = 1

This is to simulate the behavior of the system starting from any arbi-

trary profile to steady state. The entire profile should be input just as

in IFLAG = 1 above.

A sample data file is shown below and is stored in [RRK.MOD4] as MOD4.DAT.

This data files performs a simulation of the system with a step change in the

inlet gas temperature from 573° K to 623° K. The simulation is for a stop rather

than a simple disturbance since DL is set to infinity. The output will show the

simulated profiles at 10 second intervals from time 0 to 100 seconds.

0.57 30.0 P.15B7 1.194

0.25 1.041 0.005 0.274437

0.12 8.02 0.039

 -412-

2.00 8.00 75.00

17.02 0.01795 0.1436 13.09 600,0 7.163 1.25

-6.144 -48350.-2.441 10760. 17.0 0.07524

-29.44 26340. 4.385 -4615. 1.470 0.7348 6950. 18900.

0.005 100.0 0.5 9999999.

8 2000 -8 0 3 0
573.0 573.0 573.0 0.06 0.015 75.0

75.0 623.0 0.06 0.0150 0.02 0.19 0.60 573.0

LD Other Program Variables

Some of the other variables and arrays used throughout the routines are

described below. The minimum dimensions of the arrays are based on the

number of collocation points and are also described.

BWG = Biot number (wall-gas)- usually set equal to BTG

BWS = Biot number (wall-solid) -usually set equal to BTS
PGS = steady state inlet density (g/cm ..3 gas)
ASG = heat transfer area, gas-catalyst (cm..2)

AST = heat transfer area, thermal well-catalyst (em ..2)

ATG = heat transfer area, thermal well-gas (cm••2)
VT = volume of thermal well (cm..3)
VB = volume of bed (cm..3)
RG = gas constant (cm••3 atm/gmole K')
RGP = gas constant (call gmole K)
 -413-

CPG = heat capacity of gas (cal/g K••2)

MG = average molecular weight (gig mole)
The following table describes the vectors and matrices used throughout the

modeling programs.

NAME MIN. DIMENSIONS PURPOSE

A N+2,N+2 first derivative collocation matrix

AA 5"'N,5•N state matrix 'A'

AI 5"'N.5•N inverse or the state matrix

B N+2.N+2 second derivative collocation matrix

BB 5•N,8 control matrix 'B'

cc 5"'N Al"'(EX-IDENTITY)"'Q

DD 5*N constant matrix 'D'

ET 5"'N.5•N Al"'(EX-IDENTITY)

EVR 5"'N real part of eigenvalues

EV1 5•N imag part of eigenvalues

EX 5•N,5•N EXP(AA•DT)

R N+2 vector containing collocation points

RLl N.e constants for linearized methanation rate

RL2 N,9 constants for linearized steam-shift rate

SR 5•N,5*N real part of eigenvectors

SI 5*N,5•N imag part of eigenvectors

SIR 5•N.5•N real part of SR.SI inverse

SII 5•N,5•N imag part of SR,Sl inverse

u B disturbance or control vector

w 10 radial lumped coefficients

y s•N+5 state vector
 -414-

YS state vector

The structure of the state matrix (Y) and steady state state matrix (YS) is

l, ... ,N Solid temperature at collocation pts

N+1, ... ,2*N Gas temperature at collocation pts

2*N+l, ... ,3*N Thermal well temperature at collocation pts

3*N+ 1, ... ,4*N CO concentration at collocation pts

4*N+l, ... ,5*N C02 concentration at collocation pts

5*N+1, ... ,6*N+l Velocities at collocation pts plus z=l

6N+2 Solid temperature at z=O

6N+3 Solid temperature at z=l

6N+4 Gas temperature at z=O

6N +5 Gas temperature at z= 1

The structure of the control vector (U) is

1 cooling wall temperature

2 inlet velocity

3 inlet gas temperature

4 inlet CO concentration

5 inlet c~ concentration

6 inlet ~0 concentration
7 inlet H2 concentration

8 inlet Cl4 concentration

All quantities in the programs except the inputs are nondimensional. However,

the programs are hot based on deviation variables. Thus the form of the linear

model is
 -415-

X(t) = Ax(t) + Bu(t) + D

The analytic solution of this equation used by ANAIIOD, RDl.IIOD and RD2MOD is :

x(t1) =eA(~-~I(t0 ) + A- 1 [eA(~-~-I][Bu + D]

This is valid for time to to t 1 during which the control is constant.

LE Program Outputs

The program output consists of displaying the input data, the program con-

ditions, the steady state conditions, and the axial collocation points. The initial

guess for the steady state solution is then shown with the calculated steady

state solution.

The simulated profiles are then printed at the specified intervals up to the

total time period. The final steady state, calculated by setting the time deriva-

tives equal to zero in the original simulation and solving the resultant system of

algebraic equations is printed. For the linearized cases, this steady state may

be significantly in error since the model was linearized about the original steady

state and since some of the variables such as MG and CPG are based on the

steady state inputs. For this reason. the 'Actual Steady State' is also printed.

This is the steady state based on the new conditions.

Table A4-6 show the output using the above data file with TMAX = 5 for an
inlet gas temperature step change from 573° K to 623° K.

Several sets of library routines have been written for the modeling and plot-

ting programs. Some of these routines are modifications of Caltech library

 -416-

routines. In these cases, the names are unchanged. The user may want to study

the Caltech documentation that accompanies these programs. Although most

have been changed to clean up portions of the routines not needed in this work,

to reduce some o! the restrictions, and to work in double precision, the basic

techniques remain the same. These libraries are described below with brief

descriptions of the various routines in each library. All of these are found in

[RRK.lJBR].

UJB.FOR This library contains the many routines necessary for modeling the

reactor. All of these routines have been written specifically for the packed bed
model and are listed in Table A4-~f.

DIMLES calculates the dimensionless parameters.

INITIAL sets up initial profiles.

SETUPS reads the input data, makes preliminary calculations and prints

the inputs.

OUTPUT outputs the calculated profiles.

RADIAL calculates the constants for the radially lumped model.

COLLOC calculates the zeros of the orthogonal polynomial and sets up

the axial collocation matrices.

INTLSS solves for the initial steady state profile.

FN 1 defines the algebraic equations for use by NSES 1 when solving

for the steady state.

FN2 identical to FNJ but is written to be used with NSES2.

FN3 defines the steady state linearized equations for NSESJ.

 -417-

LREAC calculates the coefficients for the linearized rates.

ASETUP sets up the linearized model including the state matrix

ACTLSS calculates the final steady states.

INSIM initializes the simulation.

OUTCALC calculates the velocities and endpoint temperatures.

CPCALC calculates the gas heat capacity.

ENDPTS calculates the endpoint conditions.

REAC calculates the dimensionless rates.

ELIB.FOR This library is necessary for the calculation of the eigenproperties.

EIGENP calculates the eigenvalues and eigenvectors of a real matrix.

This routine uses the subprograms SKALE, REAL VE, HESQR,

COMPVE Lhal are also included in this library.

101B.FOR This library includes various numerical and printing routines.

DFOPR evaluates the discretization matrices for collocation. This rou-

tine is from the text by Villadsen and Michelsen.

JCOBI evaluates the roots and derivatives of Jacobi polynomials. This

subroutine is from the text by Villadsen and Michelsen.

MODDEQ may be used to solve a system of first-order differential equa-

tions.

This routine uses the subprograms MADAM, MGILL, MREST,

MSA VE that are also included in this library.

NSES1 solves a system of algebraic equations by numerically calculat-

 -418-

ing the Jacobian and using MATINVto invert it.

NSES2 solves a system of algebraic equations using Brown's quadrati-

cally convergent algorithm. This routine is slower than NSESJ

but is more powerfuL

PRINTS prints the scalars z(i) , i=l, .... , 8.

PRINTV prints the vector z(m).

PRINTM prints the matrix z(n,m).

ZEROV zeros the vector z(m).

ZEROM zeros the matrix z(n,m).

NEGCH checks a vector for any negative values. sets the negative value

to zero and prints a warning message.

MAT/NV inverts a matrix.

HNJJJNV inverts a matrix.

CINVSE inverts a complex matrix.

MATCPY copies a portion of one matrix to another.

MATMULT multiplies two matrices.

ll.A PUB.FOR

This library of routines is necessary for the plotting routines. Several of

the programs in this library are identical to those in MLIB except that they are

written in single precision for the plotting.

JCOBI same as in MUB but in single precision.

Di''UPR same as in MUB but in single precision.

 -419-

INTRP evaluates the Lagrangian interpolation coefficients.

NAXS replaces the Caltech library routine for plotting the x axis.

NYAXIS replaces the Caltech library routine for plotting the y axis.

PWI"'1NG PROGRAIIS

Several programs have been developed for plotting various types of profiles

on the Zeta plotter. These programs all must be linked to PLIB. All of the rou-

tines have slightly differing input data structures. Thus it may be necessary to

study the programs before attempting to use them the first time. Many of the

variables used by the programs are defined below.

OX,OY - location of the origin on the page

XL,XY -length of the two axes

Yl,Y2 - starting and ending values for the y axis

Xl,X2 - starting and ending values for the x axis

XT,YT - distance between tics on the two axes

ID - type of line drawn between the points

IST - 0 if the data is the last line for the plot, 1 if more lines follow

lS - speed of the plotter

N - number of collocation points

The folloWing are the most useful programs. They are in [RRK.PIDT].

PGEN plots a completely general profile.

 -420-

PLTA plots the axial temperature profiles, given the values at the col-

location points.

PLTC plots the axial concentration profile, given the values at the col-

location points.

PLTCl is identical to PLTC except that the concentration that is plot-

ted is that based on the inlet number of moles.

PLTCR plots the radial concentration profile.

PLTR plots the radial temperature profile given the values at the collo-

cation points.

PLTT plots the time profiles of the temperature using the spline rou-

tines to fit the curve.

Pf,TTC plots the concentration time profiles using the spline routines to

fit the curve.

 -421-

C THIS P~OGi~AM MODELS A IIOIUSOTHEIIMAL, IIOMAOIAIATIC FIXED lED UACTOII

C \liTH lOTH A COOLING; JACKET AND A THERMAL \fELL. THE ANALYSIS IS
C: P£1\F~MED F~ A METHANATION OF CO SYSTEM.

C LINK TO MLI8.LLI811
C THIS PROGRAMS SOLVES THE FULL MOMLIMEAR MODEL 1/ITII THE COIICEII~TIOIIS
C NONOIHENSIONAL 1/.R.T. THE STEADY-STATE INLET COliC. OF CO
IMPLICIT REAL•8 !A·M,O·ZI
lt£AL •8 l • 1<1' ,Mt:. kJI, KllP .ICP lA.KPII.ICCl •• KP2A,KP21,KC2. ,KZP ,IC3P
COMMON /REACP/ EPS,L,U,U
/CATLS/ CPS, PS, TC, DC
ITH\IEL I CPT, PT, KT
/GASP A/ CPQ, PTZ, PTII. ,Mil:, PGS, Uli:S, UM
/HEA TT I OHSG, OHTS, OHTI;, 8GS, BSGi, BTS ,ITIO, 1\/G, 1\/S
/OPCON I SCHl. SCO. SC02. SH2, SHZO. PT., PTI, ST., STII
/REACl/ DHlA,OH!B,OHZA,OHU,kJI,UP
/OlHi.E/ ALS ,ALG .AL T ,BES, BEGi ,GAS ,GiAIO ,lOTS ,;TG, OU,
DEZ, S II ,SIZ ,513 ,PHIZ, PHI3 ,PHI,PHZ ,PH3
/RADIAl \llUl,\/Pl9,2l,OETA!2l,kR
/COHAT/ AIZS,2Sl,IIZ5,ZSI,R<2Sl,N
/I'IISCI/ UUI l'*l
/R£AC2/ KC!ll. KP lA. KP II. KCZll .KPZA.KP21,K2P ,K3P, EAI,EAZ
/LINEA/ AA175,7Sl,BBI75,8l,DDI7Sl,UI8)
!STATE/ Vll.lll,YDOTIIll.l,ST,OT
EXTERNAL OUIV,FN
C READ IN DATA AND CALCULATE CONSTANTS
CALL SETUPS IOTI,TMAX,DL,M,U,EP,IFLAt;,NFLAI'O,NP,!Fl
CALCULATE THE DIMENSIONLESS PARAMETERS
CALL DIHLES
CALCULATE CONSTANn FOR THE RADIAL LUMPED MODEL
CAlL RADIAL
CALCULATE ZEROS OF THE ORTHOGONAL POLYNOMIAL AND SET UP AXIAL
COLLOCATION I'IATRICIES A AND B.
CALL COI.LOC
C SOLVE FOR THE STEADY STATE PROFILE
CALL INTLSS I Y, U,IFLAil,NFLAG ,L, Uli$, PHI,N,EP l
SPECIFY INITIAL CONDITIONS
OT•DTll
T•ll.OII
CALL INSIMNIY,U,UU,N,JFLAGl
PR !NT I
ST•UUIN+Sl
CALL NSESIIN+S,UU,EP,I.ll,I,FNJ
CALL OUTCALN lV,UU,Nl
CALL OUTPUT IY,U,T,L,UGS,PHI.lll
C IIIITIALIZE INTEGRATION ltOUT!IIE MODOEQ
LL•I
JJI•U
IIT•I
UZ
~tL MODDEQ CDEit!V ,IC,S 0 N, T ,,. ,YOOT ,DT ,£P)
SOLVE THE 5N O.D.E. 'S FROM T TO T+DT
Zlll NT•NT+ 1
CALL MOOD EO l DER IV ,K,5*N, T ,V, VDOT ,DT ,EP)
IF IK .EQ. -ll PRINTZ
IF IK ,£Q. •I l STOP
C SOLVE THE N+S ALilEBRAIC EQUATIONS AT TIME T•OT
ST•UUIN•Sl
CALL ffSESl~N+b~UU,EP,l •• ,.,FftJ

C PltlNT RESULTS AND CONTINUE PROCESS IF T IS LESS THAN TMAX

IFtLL.EQ.I .AND. T.IOE.OL) GO TO 3.1
IFtNT.NE.NP .AND. T.LT.TMAXl 1'00 TO
CALL OUTC:ALNIY,UU,Nl
z••
CALL OUTPUT lY,U,T,L,Ut;S,PHI,lll
!FIT .liE. TMAX> 100 TO •••
liT••

Table A4-1
Computer Program NLNJIOD
 -422-

C AOJUST TIME STEP IF NECESSARY

lS•lFN IYOOT,Z 0 N)
JJ•IIOINT IRSI - 3
IFIJJ .EO. 21 JJ•!IOINTilS-.2011 - 3
!FIJJ .LE. I .OR. JJ ,;[. lfl ;o TO 211
DTI•OTI•4.IOI••JJ
IFIIDTI-DTl .LE • • 111101 .AMD. IJJ1+3-RSI .LT •• IRilQO TO Ul
DT•DTI
JJI•JJ
PRINT 4, IDT•L/UGSI
1:0 TO ZIZ
C ADJUST FOR END OF IMPULSE
Ul LL•I
DO 311 1•2.8
Ul Ulll•I.DI
Ulll•PH3
PRINT 3, IT 0 L/UI:SI
DT•DT.II
1:0 TO 212
C CAlCUl.TE FIM•L STEADY STATE
Ul IFILl.EQ.I .OR. JFlAC:.EQ,JlSTOP
CALL ACTLSSIU,Y,EP,Nl
PRINT I
CALL OUTPUTIY,U,I.DI,L,U;S,PH1,2l
STOP
FORMATIIHI l
FORMAT I///,ZX,'ERROR- MODDEQ COULD NOT CONVEli:E',///1
FORMATI///////t,2X,'ACTUAl TIME OF IMPULSE END •',FI.4,///////l
FORMATI//////,2X,'NEW TIME STEP •',Fll.4//////l
END

SUIROUTINE INSIMMIV,U,UU,M,JFLAI:l
C THIS SUilOUTINE INITIALIZES TME SIMULATION.
IMPLICIT R!AL•I IA-H,0-%1
REAL•& MG,MC:I
DIMENSION U18l,YIIII1,UUIII11
COMMON /GASPA/ CPG,PTz,PTR.HG,PGS,UC:S,UM
/OPCON/ SCH4,SCO,SC02,SHZ,SHZO,PTI,PTl,STI,STV
READ I 5 ,1) JFLAG
lf<JFLAii .~Q. IICALL INITIALIY,U,Il
IFIJFLAG .EO. 11110 TO 211
READ 15,21 UGJ,TI,XCO,XCOZ,XH20,XH2,XCH4,TV
CALL CPCALCIXHZ,XCO,XCOZ,XHZO,XCH4,TI,MQI,CP,CP!,CP2l
UM•MG./MG*ICPI*TI+CP%1/CPQ
U( I l •TW/STI
UIZ1•UGI/UGS
UI 31 •TI/STI
U14l•XCO/SCO
UCSJ•XCOZISCO
Ul6l•XHZO/SCO
U17l•XH2/SCO
U18 l •XCHHSCO
PRINT 3, CUill,l•l,II,UM
lll 00 311 l•l,N+&
311 VUCIJ•YCN*6+1)
RETURN
•oRNATe 12)
FORMATI9F8.Zl
FORMATI/1/,ZX,'INITIAl CONTROL VECTOla',//SX,'TV •',FI.4,1X,
• "UGi. •• ,F8.4,SX, 'T• •",F8.4.SX. ')(CO •' ,F8.4,SX,/5)(,
'XC02•~.F8.4,5M,'KHZO••,FI.4,6K,'XH2 •',FI.4,5X,'XCH4•'•
F8.4/SX,'UM •',F8,4l
END

SUBROUTINE OUTCALN CY,UU,Nl

IMPLICIT REAL*I CA•H,O•Zl
DIMEISIOH Ylllll,UU!lll1
DO Ill I•I,M+I
Ill Y<S*N•Il•UU(IJ
UTUU
INO

Table A4-1 Continued

 -423-

SUIROUTINE O!RIY <MM,T,V,VOOTI

C THIS su•aouTIM£ DEFINI$ THE OIFERENTIAL fOUATIONS Fa« US£ IV ~0£0.

IMPLICIT REAL*8 IA·H,O·ZI

REAL*I !Ill
DIMENSION Yll •• I,YOOTII •• I
COMMON IDIMLE/ AlS,ALG,ALT,IES,IEG,GAS,GAG,GTS,GTG,0£1,
• OE2, 511, S 12, Sll, ,HI Z ,PHU ,PHI, ,HZ ,,H3
/RADIAl ~ll.J.~Pl9,2J,D£TAI2J,tR
/COMAT/ AI25,ZSI,I125,25l,RIZSI,M
IH!SCI/ UUI IWWI
/GASPA/ CPG,PTo,PTR,HG,PGS,UGS,UH
II.INEAI M175,7SI,II17S,Il,OOI751,UUI
MPZ•M•Z
NZ•M•Z
MJ•M*3
M4•M•4
TAU•PHZ·I.D.
CALL EMDPTS ITT•• YI.,Y!MPI,YZ.,YZMPI,TTMPI,Y,PHII
DO !WI J•l,M
lfl'l•l+l
PG•TAU•RIIPII+I.D•
CALL REAC IYIMJ+Il,YIN4+1l,YIJl,YIM+II,PG,RIP,RZPI
SI•SUMlY,,,I,UUIN•ZI,UUIN+Jl,I,Ml
SZ•SUMfV,~,A,UU<N•4)tUU(N+5),1,Ml
SJ•SUMIY,N,8,UUIN+41.UUIN+Sl,I,Ml
S4•SUMIY,NJ,A,YJ.,YIMPI,I,NI
S6•SUMIY,N4,A,YZ •• YZMPI,I,NJ
S8•SUMIV,N2,1,TT.,TTNPI,I,Ml
VOOTIII• ALS•SI + Wlll*Yill + W1Zl*YIN2+ll + GAS*YCM+!l
+ PE1°RIP*Il,D••,HIZ*VCill + OE2*R2P*Cl,Q.+PHI3•YIIII
+ WOI*UCI l
YDOT!N+IJ• •UUIIl*$2 + tAL~*SJ + WI.)*VIN+Il + 11151*Vt112+11
+ IOAG•VtlJ + llt6J*U!lll1Pii*YIII+IJ/UH
VDOTtN2+1J• AlT*$8 • Wt7l*Ytll + lltii*VtN+Il + Wtti•VtN2+!J
+ \1( t•J•Ut I l
YDOTtN3+[)• ·UUI[)"$4 • SI2*YtN+I1*~2P/PG • SII*VIN+II*RIP/PQ
YOOTIN4+1J• •UU([)•$6 - SIJ*YtN+II*RZP/PQ
IWI CONTINUE
RETURN
END

SUIROUTINE Fll IUU,VJ

C THIS SUIROUTIN£ 0£fiN£S THE ALCEIRAIC EOUATIONS rCA USE IV NSESZ
IMPLICIT kEAL*I tA·H,0·%1
UAL*B Mil
DIMENSION UUti •• I,Vt\1
COMHON /COMAT/ AC25,25J,BI25,2SJ,kl2fil,N
/OIMLE/ ALS,AlG,AlT,8ES,IEG,IiAS,GAii,GTS,IiTii,DEI,
DEZ.Stl,Stt,St3,~HJZ,PHI3,PH1,PH2,PH~
/STATE/ Yl 1•WJ ,.YOOTt J.JJJ,ST .DT
/LINEAl AAI75,751,88175,8J,DDt751,UI81
/GASPA/ CPG,PTr,PTR,M~,PiiS,UGS,UM
/HEATT/ OHS1i,OHTS,OHT1i,IGS,ISG,ITS,8TG,IWG,IWS
IIPZ•N+2
YCN+l)•$UMtY •• ,A,UU(M+2l,UUlM+3)••·"••1&S~(UUlN+tl~UU(••2>)
VIN+3l•SUMCY,.,A,UUIN+Zl,UUIN+JI,N+I,III+IGS*IUUtN+31·UUIN•51J
YCN••I•SUMIY,N,A,UUIN••I,UUIN+5J,I,III+BSG*IUUIN+2l·UU!N+4Jl
+UH*Ut21•tUI3l·UUtN+4ll/Alli/UUIN+.l
YtN+5l•SUHIY,N,A,UUtM+4l,UUtN+SI,N+I,IIl+ISii*IUUIN+5l-UUIN+lll
TAU•PHl-1.0.
z•• DO Ul K•I,N
VI K I •Vt N>K I •SUM! UU •• ,A, Ul 2 I, UUI N+l l, K, Nl-UU( K l *SUM IV, N ,A,
UUIN+.l,UU!N+5!,K,NI·YDOTtN+Kl+UUC<I*Y!N+Kl*TAU/
ITAU•R t K+l l+l.Oil
VCII+I I•UUU+6 I *SUM! UU,.,A. Ul Z l, UUIII+I l, II+ I, MI·UUIN+I I*SUMI Y,II,A,
~¥in::~~~IN+Sl,N•I,NI-IUU111+5l•5TJIOT>UUIN+Il*UUtii+5J*TAU/
RETURN
UD

Table A4-1 Continued

 -424-

C THIS ,ROGRAM MODELS A NONISOTHERMAL, NONADIABATIC Fl~ED BED REACTOR

C . WITH BOTH A COOLING JACKET AND A THERMAL WELL. THE ANALYSIS IS
C PERFORMED FOR A METHANATION OF CO SYSTEM.
C THIS PROGRAM PERFORMS THE SIMULATIONS FOR THE LINEARIZED MODEL
C USING THE NUMERICAL DIFFERENTIAL EQUATION SOLVER MODDEQ.
c
C LINK TO MLIB,LLIB
IMPliCIT REAl•B IA-H,O-Zl
REAL*8 L,KT,MG,KI,KIP,KPlA,KPlB,KCl.,KP2A,KP28,KC2B,K2P,K3P
DIMENSION YSII •• l,RL1115,9l,RL2115,9l,YII.Bl,YDOTI1BBI
COMMON /REACP/ EPS,L,R.IJ,Rl
• /CATLS/ CPS,PS,TC,DC
• /THWEL/ CPT,PT,KT
• /GASPA/ CPG,PTZ,PTR,MG,PGS,UGS,UM
• /HEATT/ OHSG,OHTS,OHTG,BGS,8SG,8TS,BTG,BWG,BWS
• /OPCON/ SCH4,SCO,SCOZ,SH2,SH20,PTB,PTI,ST.IJ,STW
/REACI/ DHIA,DHIB,DHZA,OHZB,KB,KBP
/OIM~E/ ALS,ALG,ALT,BES,BEG,GAS,GAG,GTS,GTG,DEl,
• DEZ,SII,SI2,SI3,PHIZ,PHI3,PHI,PH2,PH3
• /RADIAl WII.IYl,WP19,2l,OETAI2l,RR
• /COMAT/ A(25,251,8(25,25l,RIZ51,N
• /REAC2/ KCl.IY,KPIA,KPIB,KC2.1Y,KP2A,KP28,KZP,K3P,EAl,EAZ
• /LINEA/ AAI75,751,BBI75,8l,DDI751,UIBI
* /LALGB/ SC4,41,HI2B,4.1Yl,RII2B,4l,ALI7l
EXTERNAL FN3,DERIV
C READ IN DATA AND CALCULATE CONSTANTS
CALL SETUPS IDTB,TMAX,DL,N,RR,EP,IFLAG,NFLAG,NP,IFl
C CALCULATE THE DIMENSIONLESS PARAMETERS
CALL DIMLES
C CALCULATE CONSTANTS FOR THE RADIAL LUMPED MODEL
CALL RADIAL
C CALCULATE ZEROS OF THE ORTHOGONAL POLYNOMIAL AND SET UP AXIAL
C COL~OCATION MATRICIES A AND 8.

CALL COLLOC
C SOLVE FOR THE STEADY STATE PROFILES
CALL INTLSSIYS,U,IFLAG,NFLAG,L,UGS,PHI,N,EPl
C CALCULATE THE COEFFICIENTS FOR THE LINEARIZED REACTION RATES
CALL LREACIYS,RLI,RL2l
CAll ASETUPIYS,Rll,Rl21
C SPECIFY THE PROFILE AT T•B
T•B.D.IY
CALL INSIMIY,YS,U,N,JFLAGI
PRINT 1
CALL OUTCALCIU,Y,YSIN*6•4l,ALGI
CALL OUTPUTIV,U,T,L,UGS,,H1 •• l

LL•B
JJ1•2.8'
NT•.f
DT•DT.f
z•2 K•t
CALL MOODEOIDERIV,K,S•N,T,V,VDOT,DT,EPI
C SOLVE THE SN O.D.E.'S FROM T TO T+DT
ZBB NT•NT+l
CALL MODDEQIDERlV.K,&•N,T,V,YDOT,DT,EPI
H!K .EO. -1 l PRINT 2
IFIK .EQ. -11 STOP

Table A1-2
Computer Program LIN][OD
 -425-

C ,~tNT ~ESULTS AND CONTINUE IF T < TMAX

IF!LL.EQ .•• AND. T.GE.DLI GO TO 3••
IFINT.NE.NP .AltO. T.LT.TMAXI GO TO 2BB
CALL OUTCALCCU,V,VS(N*6+4l,AlGl
CALL OUTPUT!V,U,T,L,UGS,PHI,BI
IF!T .GE. TMAXJ GO TO 4BB
IIT•B
C ADJUST TIME STEP IF NECESSARY
RS•RFII CVDOT,Z•NJ
JJ•IIOINTIRSI-3
IF<JJ .EQ. 21 JJ•IIDJNT!RS-.ZDBI • 3
IFIJJ.LE.B .OR. JJ.GE.IFlGO TO ZBB
DTl•DTB*4.0B**JJ
IFC!DTI-DTJ .LE . • BBB!DB .AND. (JJt+3-~$) .LT •• IOBH~O TO ZBB
DT•DTl
JJl•JJ
PRINT 4, !DT*L/UGSI
GO TO ZBZ
C ADJUST FOR END OF IMPULSE
3BB ll•l
DO 361 1•2,8
3B1 U!II•l.DB
Utii•PH3
PRINT 3, !T*L/UGSI
DT•DTB
GO TO 2B2
C CALCULATE FINAL STEADY STATE
UB IFILL .EQ. l .011.. JFLAG .EQ. l>STOI'
PRINT 1
CALL NSESI!S*N,V,EP*I.D-2,1BB,B,FN31
CALL OUTCALCIU,Y,VSCN*6+4l,ALGI
CALL OUTPUT!V.U.B.DB.L,UGS,P~1 ,21
CALL ACTLSS!U,V,EP,NJ
CALL OUTPUT !Y,U,B.D8,L,UGS,PH1,31
STOP
I FORMAT!lHII
2 FORHATC///,2X,'ERROR- MOOOEQ COULD NOT CONVERGE",///)
3 FORMAT!//1/////,ZX,'ACTUAL TIME OF IMPULSE END •',F8.4///////I
4 FORMAT!////1///,ZX,'NEW TIME STEP •',F1B.4///////l
END

SUBROUTINE DERIV IN,T,Y,YDOTI

C THIS SUBROUTINE DEFINES THE DIFFERENTIAL EQUATIONS FOR USE SV MODOEQ
IMPLICIT REAL*8 IA-H,0-%1
DIMENSION Y!lBBI,YOOTClBBI
COMMON /LINEA/ AA175,751,BBI75,81,DOI751,UI81
• /LALGB/ SI4,41,H!ZB,A.l,Rl128',Al,All71
00 18'8' I•I,N
YDOTII >•DOll I
DO lBl J•l.N
111 YDOT!li•YDOT!li+AAI!,J)•V!J)
DO 18'. J•l, 8
IBB YDOT!II•YDOT!ll+II!I,JI*U!JI
RETURN
EIID

Table A4-2 Continued

 -426-

C THIS rRO~RAM MODELS A MOMISOTHERMAL, MOMAOIABATIC FIXED lEO REACTO«

C WITH 80TH A COOLIN~ JACKET AND A THERMAL WELL. THE ANALYSIS IS
C 'ERFORMED FOR A METHANATION OF CO SYSTEM.
C THIS PROGRAM PERFORMS THE SIMULATIONS FOR THE LINEARIZED MODEL
C USING THE ANALYTIC SOLUTION.
C LINK TO MLIB.LLIB,ELIB
IMPLICIT REAL*& IA-H,O-Zl
REAL*8 L,KT,H~,KB,KBP,KPIA,KP18,KCII,KP2A,KP28,KC2B,K2P,K3P
DIMENSION VSillll,RLII15,91,RL2115,9),VIllll,YDOTIIIBl
COMMON /REACP/ EPS,L,RB,Rl
* /CATLS/ CPS,PS,TC,OC
/THWEl/ CPT,PT,KT
* /GASPA/ CPG,PTZ,PTR,MG,P~S.UGS,UM
* /HEATT/ OHSG,OHTS,OHTG,BGS,BSG,BTS,BTG,BWG.BWS
/OPCON/ SCH,,SCO.SCOZ,SHZ,SHZO,PTI,PTI,STB,STW
• /REACt/ OHJA,DHlB,OH2A,OH2B,KB,KBP
* /OIMLE/ ALS,ALG,ALT,BES,BEG,GAS,GAG,GTS,GTG,OEl,
* OE2.Sil,Sl2.SI3.PHJ2,PH13,PHl,PHZ,PH3
* /RADIAl Wlllfl,\IPI9,2l,OETAI21,RR
/COMAT/ A12S,251,8125,2Sl,RI25l,N
/REAC2/ KCIB,KPIA,KP\B,KC28,KP2A,KP2B,KZP,K3P,EA1,EA2
• /LINEA/ AAI 75,751,88175,81,001751 ,UI8l
• /LALGB/ SI •. 4J,HI2S .• Bl,RII2B,4l,All7l
* /EIGEN/ EVRI75l,EVI175l,SRI75,75l,SII75.7SJ,SIRI75,751,
• Sll175,75l
/ANLVT/ A1175,75l,EXI75,75l,ETI75,75l,CCI76l
ElCTERNAL FN3
C READ IN DATA AND CALCULATE CONSTANTS
CALL SETUPS IOT,TMAX,OL,N,RR,EP,IFLAG,NFLAG,NP,IF>
DT•DT*NP
C CALCULATE THE DIMENSIONLESS PARAMETERS
CALL OIMLES
C CALCULATE CONSTANTS FOR THE RADIAL LUMPED MODEL
CALL RADIAL
C CALCULATE ZEROS OF THE ORTHOGONAL POLYNOMIAL AND SET UP AXIAL
C COLLOCATION MATRICIES A AND B.
CALL COLLOC
C SOLVE FOR THE STEADY STATE PROFILES
CALL INTLSSIVS,U,IFLAG,NFLAG,L,UGS,PHI,N,EP>
C CALCULATE THE COEFFICIENTS fOR THE LINEARIZED REACTION RATES
CALL LREACIVS,RL1,RL2l
CALL ASETUPIVS,RLI,RLZ>
C SPECIFY THE PROFILE AT T••
T••.D•
CALL INSIMIV,VS,U,N,JFLAGl
I"RINT 1
CALL OUTCALCIU,V,VSIN*6+,),ALG>

Table A4-3
Computer Program ANAIIOD
 -427-

CALL OUT,UTIY,U,T,L,U,S,,Hl,ll
C CALCULATE THE ANALYTIC CONSTANTS
LL•8'
IN•S'
212 CALL ANALYTIDT,N,INI
C SOLVE THE 5N O.D.E.'S FROM T TO T+DT
z•g DO z•3 l•l,&*N
YOOT!Il•CC!Il
DO 2.1J3 J•l,S*N
2.3 VDOTIIl•YOOTIIl+EXti,J)*YIJI
T•T+OT
DO ZJJ! l•l,S*N
211 Ylli•YDUTill
C ~RINT RESULTS AND CONTINUE IF T < TMAX
IF!Ll.EQ.8' .AND. T.GE.DLl GO TO 3••
CALL OUTCALCIU,V,YSIN*6+4l,ALGl
CALL OUTPUTIV,U,T,L,UGS,PHl,ll
!FIT .GE. TMAXl GO TO 41.1J
GO TO 2/11
C ADJUST FOR END OF IMPULSE
31. LL•l
DO 3.1 1•2,8
3.1 Ulll•l.OS'
Ul I l•PH3
PRINT 2, IT*L/UGSI
IN•Z
GO TO 28'2
C CALCULATE FINAL STEADY STATE
U8' IFILL .EQ. 1 .OR. JFLAG .EO. liSTOP
PRINT 1
CALL NSES115*N,Y,EP*1.D-1,111,1,FN3)
CALL OUTCALCIU,Y,VSIN*6+4l,ALGl
CALL OUTPUTIY,U,I.DS',L,UGS,PH1,21
CALL ACTLSS!U,Y,EP,Nl
CALL OUTPUT IY,U, •. D.,L,UGS,PHI,3l
STOP
1 FORMATIIHII
2 FORMATI/1//////,ZX,'ACTUAL TIME OF IM,ULSE END •',FI.4///////)
END

Table A4-3 Continued

 SUJ~OUTINE ANALYT <OT,N,INl

C THIS SUBROUTINE CALCULATES THE CONSTANTS NECESSARY ,Ok TM£

C ANALYTICAL SOLUTION.
IMPLICIT REAL•8 CA-H,O-Zl
DIMENSION 1HDC75>,QI75>
COMMON /LINEA/ AAI75,75>,11175,8l,DDC75l,Utll
/El~EN/ EVRI75l,EVlC75l,SRI75,75l,Sl175,751,SIR<7S,11),
Sl I 175,751
/AN~VT/ Al175,7S>.EXt75,75J,ETC75,75l,CCt75l

lFI!H .EO.
l~l!N
11 GO TO
.EO. 21 GO TO 3S6
z••
C CAlCULATE THE EIGENPROPERTIES
CALL Z£ROMISR,75,75J
CALL ZEROMI$!,75,75>
CALL 2EROVCEVR,75>
CALL ZEROVIEVI,75>
CALL MATCPVIAA,Al,75,75,75,?5,75,75l
CALL El~EHPI5*N,?S,Al,EVR,EVI,SR,SZ,lNDl

C CALCULATE S INVERSE AND A INVERSE

CALL ZEROMISIR,7S,7S>
CALL 2EROMISII,75,7Sl
CALL MATCPYIAA,Al,7S,75,75,75,75,751
00 11616 I•1.N*5
11.1 S 1R I I , 1 l • I • Dl
CALL CINVSE!SR.SI,SIR.Sl1.7S,S*Hl
CALL 8NOINVIA!,S*N,lTEST>
C CALCULATE EXP!LAMBOA*DT>
ZIB 00 261 I•t,S•H
CC{I>•D,XP!EVACl)*DT>•DC0$1EVICll*DT>
211 Oll>•OEXPIEVRII>*DT>*DSlHIEVllli"'DTI
DO 263 1•1,5*11
00 2113 J•I,S"'N
STR•16.016
STI•16.016
DO 2162 K•I,S*N
ST1•16.0ll'
ST2•16. 016
CALL CMULTICC/KI,OIK!,SlRIK,Jl,S111K,J),STl,ST21
zgz CALL CMULT!SR/l,K>,Slll,Kl,STI,STZ,STR,STll
213 EXIl,Jl•STR
C CALCULATE ADDITIONAl CONSTANTS
DO 2164 l .. l,S*H
DO Z.IU J•I,S*N
ETtl.Jl•JI.OI
DO 2S.t K•l,S•N
ALP•EXIK,Jl
IFIK .EO. JIALP•ALP•I.OJI
214 ETil,Jl•ETII,JI•Alii,Kl*ALP
3BB DO 3161 l•t,S•N
01 I l•Dtll I l
00 361 J•l,8
311 Olll•OIII~BB/l,J>*UIJ>
DO 362 I•t,S•II
CC(f , .... 0.
00 3.2 J•t,S•III
l•z CCIII•CClii•Ei([,JJ•QCJJ
RETURN
END

Table A4-3 Continued

 -429-

C THIS PRQ;RAM MODELS A NONISOTHERMAL, NONAOIAIATIC FIXED lEO REACTOa

C WITH lOTH A COOLIN; JACKET AND A THERMAL WELL. TM£ ANALYSIS IS
C PERFORMED FOR A METMANATION OF CO SYSTEM.
C THIS PRQ;RAM PERFORMS THE SIMULATIONS FOR THE REDUCED LINEAR MODEL
C USIN; THE ANALYTIC SOLUTION. THE ASSUMPTION OF NEGLIGIBLE ENERGY
C ACCUMULATION IN THE GAS AND QUASI SS FOR CONCENTRATION ARE US£0
C FOR THE MODEL REDUCTION.
C LINK TO MLIB,LLII.ELIB
IMPLICIT REAL•& IA-H,O-Zl
R£AL"8 L,KT,HG,K.,K.P,KPIA,KPII,KCI.,KPZA,KPZI.KC2 •• K2P,K~P
DIMENSION YSI " " ,RL I I 15,91 ,RL21 15,91, VI 1•• ,, VOOTI 1••1
COMMON /REACP/ EPS,L,R •• RI
/CATLS/ CPS,PS,TC,OC
/THVEL/ CPT,PT,KT
• /GASPA/ CPG.PTZ,PTR,MG,PGS,UGS.UM
• /HEATT/ OHSG,OHTS,OHTG,BGS,ISG,BTS,ITG,IVG,IVS
/OPCONf 5CH4.SCO~SCOZ.$~l,SHZO.,T._tTl.IT •• 3TW
/REACI/ DHIA,DHIB,DH2A,DHZB,~,K·P
/DIMLE/ ALS,ALG,ALT,BES,BEG,GAS,GAG,GTS,GTG,OE!,
DEZ,SII,SIZ,SI3,PHI2,PHI3,PHI,PHZ,PH3
/RADIAl IIII.>,VPI9,21,DETAI21,RR
/COMAT/ Al25,25l,BI25,ZSI,RI251,N
/REACZ/ KCI •• KPIA,KPIB,KC2.,KPZA,KP21,K2P,K3P.EAI,lAZ
/LINEA/ AAI75,75l,BBI75,8l,D01751,U181
z•
/LALGB/ 514,4 I ,HI .4.1 ,R I I z•.4 I ,AL!7 I
/EI,EN/ EVRI7SI,EVII75l,SRI7&,751,SI(75,751,S1Rt76,711,
Sli!75,7SI
/ANLYT/ AI175,75>,EX175,751,ET175,7SI,CCI7&1
/REDLN/ ACI 75,75 I ,API 75,75 I ,QPI 75 l ,CI I 75,75 I ,C2!75,11,
AC H 75,76 l

EXTERNAL FN3
C READ IN DATA AND CALCULATE CONSTANTS
CALL SETUPS !OT,TMAX,OL,N,RR,EP,IFLA;,NFLAG,NP,Ifl
l.H•OT•NP

C CALCULATE THE DIMENSIONLESS PARAMETERS

CALL DIMLES
C CALCULATE CONSTANTS FOR THE RADIAL LUMPED NODEL
CALL RADIAL
C CALCULATE ZEROS OF THE ORTHQ;ONAL POLYNOMIAL AND SET UP AXIAL
C COLLOCATION MATRICIES A AND 8.
CALL COLLOC

C SOLVE FOR THE STEADY STATE PROFILES

CALL INTLSSIYS,U,IFLAG,NFLAG,L,UGS,PHI,N,EPI
CALCULATE THE COEFFICIENTS FOR THE LINEARIZED REACTION RATES
CALL LREACIYS,RLI,RLZI
CALL ASETUPIYS,RLI,RL2l
C SPECIFY THE PROFILE AT T•l

r-•.o•
CALL INSIMIY,YS,U,N,~FLAGl
PRINT I
CALL OUTCALCIU,Y,YSIN•6+4l,ALGI
CALL OUTPUT!Y,U,T,L,UGS,PHI,.I
C CALCULATE THE ANALYTIC CONSTANTS
LL••
IN••
212 CALL REDUCEIN,INI
CALL ANALYTIDT,N,INl
C SOLVE THE &N O.D.£.'S FROM T TO T+DT
z•• DO 2•3 I•I,N
YDOn 1 >•ccc 11
YDOTII+Nl•CCIN+ll
DO 21!'3 J•I,N
YOOTIN•II•VOOTIN+II+EXII+N,Jl•YI~I+EXII+N,J+Nl•Yt2*N+~)
t•3 YOOTtl}•VOOTCl>•EKtl,J>*V(J)+EX<19J•N>•Vt2*M•J)
T•T•DT
DO 2•1 1•1. N
V!Z"N•II•YDOTIN+Il
Zll YIII•YDOTIIl
C PRINT RESULTS AND CONTINUE IF T < TMAX
IFILL.EO ••• AND. T.GE.DLl GO TO ~·
CALL OUTRAIY,NI
CALL OUTCALC!U,Y,YSIN*I+41,ALCl
CALL OUTPUTIY,U,T,L,U;S,PHI,II
IF!T .;E. TMAXI GO TO 4••
C:O TO 211

Table A4-4
Computer Program RDlliOD
 -430-

C ADJUST YO. END OF IMPULSE

.,.. LL•I
DO 3.1 1•2,8
3.1 Ullla!.D.
Ull!•PH3
PRINT I, IT 0 L/UC:Sl
IN•2
1:0 TO UZ
C CALCULATE FINAL STEADY STATE
••• HILL ,!Q. I ,Ofl,
PRINT I
~fLAG ,fQ, llliTOP
CALL NSESIIS•N,Y,EP•t.D-2.1 •••• ,FN3l
CALL OVTCALCIU,Y,YSIN*&+4l,ALGl
CALL OVTRAIV,Nl
CALL OUTPUTIY,U, •. D.,L,UGS,PHI,Zl
CALL ACTLSSIU,Y,EP,Nl
~;~~ OUTPUT tY,U, •. O.~L•U~S,PH1,3l

1 FORMAT I IHI l
2 ~~HATIIIIIIIII,ZX,'ACTUAL TIME OF IMPULSE END •',FI.4/////f/l

SUIROUTINE ANALYT IDT,N,IIl

C THIS SUBROUTINE CALCULATES THE CONSTANTS NECESSARY ,Ga TKE
C ANALYTICAL SOLUTION.
IMPLICIT REAL*8 IA-H,O-Zl
DIMENSION INDI751,0175l
COMMON /LINEA/ AA175,75l,l8175,8l,DDI7Sl,UI8l
tEIGEN/ EVRI75l,EVI175l,SRI75,75l,SII75,75l,SIRI71,71l,
SIII7S, 75l
/ANLYT/ AII75,7Sl,EX175,75l,ET175,7Sl,CCI75l
/RtDLN/ ACC 75, 7Sl .Ar ( 7S, 7$) ,QPI ?& > tCl ( 7&, 75) ,C2C'71.,1).
AC1175, 751
!FilM .EO. I l GO TO Uf
IFIIN .EQ, 2> GO TO 311
C CALCULATE THE EIGENPROPERTIES
CALL ZEROMISR,75,7Sl
CALL ZEROMISI,75,7Sl
CALL ZEROVIEVR, 75l
CALL ZEROVIEVI,75l
CALL MATCPYIAP.AI,7S,7S,75,71,71,75l
CALL EIGENPI2*N,7S,AI,EVR,EVI,SR,SI,INDl
C CALCULATE S INVERSE AND A INVERSE
CALL ZEROMIS!R,75,75l
CALl ZEROM!SI1,75,75l
CALL MATCPYIAP,A1,75,75,7S,75,75,75l
DO 1•.11 I•I~N•l
llJlJ SIRI!,I)•l.DI
CALL ClNVSEISR.SI,$1R,SII,75,Z*Nl
CALL BNOINVIAI,Z*N.ITESTl
C CALCULATE EXPILAMBDA*DTl
z•• DO Zll l•192*N
CCIIl•DEXPIEVRill*DTl*DCOSIEVIIIl 0 DTl
ZlJI Ql l l •DEXP I EVR I I l*DTl*DS INI EVIl I >•DT>
DO ZlJ3 1•1,2*N
DO ZlJ3 J•I,2*N
STR•lJ.DlJ
STl•lJ.DI
00 ZlfZ K•l,Z*N
STI•lJ. DB
STZ•I.DI
CALL CMULTICCIKl,OIKl,SIRIK,Jl,SIIIK,Jl,STl,STZl
ZlJ2 CALL CMULTISRII,Kl,SIII,Kl,STI,ST2,STR,STil
213 EXII,Jl•STR
C CALCULATE ADDITIONAL CONSTANTS
DO 214 1•1.2"N
DO ZB4 J•I,2*N
ETII,Jl•I.DlJ
DO ZlJA K•I.Z*N
ALP•rX(K,.l}
IF!K .EO. JlALP•ALP-l.DlJ
2lJ4 ET!I,Jl•ETII,Jl+AIIl,Kl*ALP
31lJ DO 3lJ2 I•I,Z*N
CCII l•lJ.OlJ
DO 3lJZ J•I,Z*N
JlJZ CC!l!•CC!Il•!T!I,~l*QPI~l

~ETUIIN
END

Table A4-4 Continued

 -431-

c
C THIS SUB~OUTINE PE~FORMS THE PHYSICAL MODEL REDUCTION FOR
C USE VITH THE ANALYTIC SOLUTION.
c
IMPLICIT REAL"8 IA-H,O-Zl
DIMENSION TliSS,SSl,T215.,5Sl
COMMON /LINEA/ AA175,75l,88175,8!,00175l,UI8l
/REDLN/ ACI75,75l,AP!75,7&l,QPI7Sl,CII75,75!,C2175,1),
ACI 175, 75!
IFIIN .EQ. Zl ;o TO z••
"l:•l*M
H3•3*N
N4•.t•N
DO 1•11 I•l,N
DOIII•J•l.N
ACII.Jl•AAIN+I,N+Jl
ACII,N•Jl•AAIN+I,N3+Jl
AC I I ,NZ+J l•AAI N+I,NA+J l
ACIN+I,Jl•AAIN3+1,N+Jl
ACIN+I,N+Jl•AAIN3+1,N3+Jl
ACIN+I,NZ•Jl•AAINJ+I,N~+Jl
ACIN2•l.Jl•AAIN4•1,N+Jl
AC<N2•1,N+Jl•AAIN~+I,N3+Jl
ACfN1+1.N2+Jl•AAIH,+t.N,+J)
Tl!l,Jl•AA!I,N+Jl
T1 I I,N•J l•AA< I,N3+J l
Tl I I,NZ+Jl•AA< I,N4+Jl
Til N+l ,J l•AAI NZ+I, N+J l
T1 IN+l,N+J l•AAINZ+I,N3+Jl
T I I N•1, NZ+J l•AAI NZ+I, U+J)
T2(l,.J)•AAtN•I.J)

..
TZI I,N+J l•AAIN+I,NZ+Jl
TZ I N+1 ,J l•AAI N3+1 ,J l
TZIN+I,N+Jl•AAIN3+1,NZ+J)
, TZ!NZ+I,J l•AAIU+l,J l
TZ I N2•1, N+J) •AAI U+l, N2+J l
CALL MATCPVIT2,C1,5 •• 5/6,75,75,N3,N2l
CALL BNOINVCAC,Nl.JTtST)
CALL HATCPVIAC.ACI,75,75,75,75,N3,N3l
CALL MATMULTIAC,C1,75,7S,7S,75,N3,N2,N3l
CALL MATMULTIT1,AC,SS,SB,75,7S,N2,N3,N3l
CALL MATMULTIAC,T2,75,7S,S •• S•• NZ,NZ,N3l
DO 1•1 I•I.N
OU 181 J*l,Pt
API I,J l•AA! I,J l·TZI I,Jl
APII,N+Jl•AA<I,NZ+Jl-T211,N+J)
AP!N+I,Jl•AA!NZ+I,Jl•TZIN+I,Jl
APIN+I,N+Jl•AA!N2+l,NZ+Jl·T21N+I,N+Jl

~~~~~f,!ob!=·l>
TZIN+I,Il•DDIN3+1l
TZINZ+l ,1 l•DDIN4+1 l
Tl I I, ll•DDI ll
Tl!N+I,Il•DDIMZ+Il
DO 1•2 J•l,l
T2tl.ll•T2rt.ll+aRt••I.~l*UIJ)

TZIN+l,ll•TZIN•l,ll+BBIN3+1,Jl*U1Jl
TZINZ+I,Il•TZINZ+I,Il+IIIN~+I,Jl*UI~l
T1 I I , I l •T Ill, I l +181 I, J l •u I J l
TI!N+I,Il•TICN+I,Il+IICN2+1,Jl*UCJI
CALL MATCPYITZ,C2,S.,S.,75,l,N3,Jl
CALL HAiHULT(ACI,CZ.75,75,75.1,H3,1.H3)
CALL MATMULTIAC,TZ,75,75,S/6,5S,NZ,I,N3l
DO 1•3 I•I,NZ
QP! I l •TI I I, I l -TZ I I, I l
RETURN
END

SUBROUTINE OUTRAIY,Nl
IMPLICIT REAL•B IA-H,O-ZI
DIMENSION Yll •• l
COMMON /REOLN/ ACI75,75l,APC75,7Sl,QPI75l,Cll75,75l,CZI71,11,
.I.C1!7li. 76!

NZ•Z*N
N3•3"N
DO t••
N.4•4*M
l•I,N
VI I+Nl•-CZI 1,1 l
Vtl+N3l•-C:(H+J,1)
VII+N4l•-CZINZ+I,ll
DO 1. .J•I,N
YIN+Il•YIN+ll·CIII,Jl*YIJl•Clii,N+Jl*YINZ+J)
t•• YIN3+1>•YIM3+Il•CIIN+l,Jl*YIJl•CICM+I,M+Jl*YIM2+Jl
VUA+l >•YI N4•1 l•CII NZ•I,J I*YIJ) •CIIN2+ I, N+J l*Y! IIZ+J l
IETUU
£NO

Table A4-4 Continued

 -432-

C TMIS PROGRAM ~OELS A MOMISOTHERMAL, MONAOIAIATIC FIXED 1£0 REACTOR

C VITH lOTH A COOLING ~ACKET AND A THERMAL WEll. THE ANALYSIS IS
C PERFORMED FOR A METMAMATION OF CO SYSTEM.
C THIS PRO;RAM PERFORM$ THE SIMULATIONS FOR THE REDUCED LINEAR MODEL
C USIN~ THE ANALYTIC SOLUTION. THE ASSUMPTION OF QUASI SS FOR
C CONCENTRATION IS USED FOR THE MODEL REDUCTION.
C LINK TO MLIB,LLII,iLIB
IMPLICIT REAL*B IA-H,O-Zl
REAL*8 L,~T,MG,KW,KIP,~PIA,KPIB,KCI.,KPZA,KPZB.KCZ•.KZP,KJP
DIMENSION VS I 1661 ,RLI I 15, 9l,RLZ I 15,9 J, VI 1••>, YDOTI 1••1
COMMON fREACPf EPS,L,R.,RI
fCATLSf CPS,PS,TC,DC
tTHWEL/ CPT,PT,KT
fGASPAf CP;,PTZ,PTR,M;,p;s,u;S,UM
fHEATTf OHS;,OHTS,OHTG.B;S,BSG,BTS,ITC,I~.IWS
/OPCONf SCH4,SCO,SCOZ,SHZ,SHZO,PT6,PT!,ST•• STV
/REACI/ OHIA,DHIB,OHZA,DHZB,K.f,KIP
fDIMLEf ALS,ALG.ALT.BES.BEG,GAS,GAC,CTS,CTC,D£1,
3
/RADIAl 8f~-~~~P~i~z~~~i~~~~j~:~ ,PHI,PHZ,PH
3

/COHAT/ A125,251,81ZS,2SJ,RI25l,N
• /REACZf KCI6,KP1A,KPIB,KCZ6,KP2A,KP21,KZP,~3P,£Al,IA2
• /LINEAl AA175,7Sl,88175,8l,ODI75l,UI8l
fLAl~B/ Sl 4,4) ,HI 26 ,46l,R I I 26,4 l ,ALI7l
fEIGENf EVR17Sl,EV1175J,SRI75,7SJ,SII75,75J,Sial71,75l,
Sll I 75,75)
/ANLYT/ AI175,7Sl,EX175,75J,ET176,75l,CCI76l
/REDLN/ ACI75,7SJ,AP175,75l,QPI75l,C1175,75l,C2171,1l,
ACII75,7SJ
EXTERNAL FN3
C READ IN DATA AMO CALCUL•T• CONSTANTS
CALL SETUPS IDT,TMAX,DL,N,RR,EP,IFLAG,NFLAG,MP,IFl
DT•OT•NP
C CALCULATE THE DIMENSIONLESS PARAMETERS

C CALCULATE CONSTANTS FOR THE RADIAL LUMPED MODEL

CALL RADIAL
C CALCULATE ZEROS OF THE ORTHOGONAL POLYNOMIAL AND SET UP AXIAL
C COLLOCATION ~TkiCI'S A AHO 8.
CALL COLLOC
C SOLVE FOR THE STEADY STATE PROFILES
CALL INTLSSIYS,U,IFLAG,NFLAG,l,UGS,PHI,N,EPJ
C CALCULATE THE COEFFICIENTS FOR THE LINEARIZED REACTION RATES
CALL LREACIVS,RL!,RLZl
CALL ASETUPIYS,RLI,RlZJ
C SPECIFY THE PROFILE AT T••
T••.o•
CALL INSIHIV,YS,U,N,JFLAGJ
PltlNT I
CALL OUTCALCIU,Y,YSIN*6+4l,ALGJ
CALL OUTPVTIY,U,T,L,UGS,PHl,.l
C CALCULATE THE ANALYTIC CONSTANTS
LL•6
IN•6
z•z CALL REDUCEIN.!Nl
CALL ANALVTIDT,N,lNl
C SOLVE THE SN O.O.E.'S FROM T TO T+DT
z•g 00 263 1•1,3*N
VOOTIIl•CCIIJ
DO 2.1J3 J•1,3"N
z•l VOOTIIl•VOOTIIl+EXII,Jl*YIJJ
T•T•DT
DO 261 1•1,3*N
z•t VI ll•VOOTIIl
C PRINT RESULTS AND CONTINUE IF T < TMAX
lrtLL.EQ .•• AND. T.QE.OLl QO TO a.•
CALL OUTRAIY,Nl
CALL OUTCALCIU,Y,VSIN*'+4l,ALGl
CALl OUTPUTIV,V,T,L,UGS,PHl,.l
!FIT .GE. TMAXl CO TO 4••
GO TO 2. .

Table A4-5
Computer Program RD2110D
 -433-

C ADJUST FOR END OF IMPULSE

,.. Ll•l
DO 3.1 1•2,1
s•1 U!ll•I.D.
Ulll•PH3
PRI N1 2, IT•LJUCSI
IN•2
~:o TO z•2

...
C CALCULATE FINAL STEADY STATE
IFILL .EO. I .OR. JFLAI: ,£Q. IISTOP
PRINT I
CALL NSESII&*N,Y,EP*I.O•Z,I •• , •• FN31
CALL OUTCALCIU,Y,YSIN*6+4l,ALGI
CALL OUTRAIY,NI
CALL OUTPUTIY,U, •. O.,L,UGS,PHI,2l
CALl ACTLSS(U.Y,EP,")
CALL OUTPUT IY,U ••• D•• L.UQS,PH1,3l
STOP
I FORMAT I IHI l
2 ~~MATI////////,2X,'ACTUAL TIME OF IMPULSE END •',FI.&///////1

SUI.OUTINE ~YT IDT,N,INl

C THIS SUBROUTINE CALCULATES THE COWSTANTS NECESSAAY FOR THE
C ANALYTICAL SOLUTION.
IMPLICIT REAL*8 IA-H,O-Zl
DIMENSION IND175l,QI76l
COMMON /LINEA/ AAI75,75l,8817S,BI,DDI7SI,UIIl
/EIGEN/ EVR!751,EVI1751,SRI75,7SI,SI17S,751,SIRI71,711
$llt7S,76J
• /ANLYT/ AI175,7SI,EX175,7Sl,ET175,7Sl,CCI751
/REOLN/ AC! 75,75 I ,API 75,75 I ,QPI751 ,Cll15, 761 ,t2111,1 I,
ACJ( 75,75 I
IFI IN .EO. I l GO TO 2••
IFIIN .EQ. 21 GO TO 3.B
C CALCULATE THE EIGENPROPERTIES
CALL ZEROHISR,75,751
CALL ZEROHI$!,75,751
CALL ZEROVIEVR,7SI
CALL ZEROVIEVI,75l
CALL HATCPVIAP,AI,75,75,7~,76,75,7tl
CALL £1;ENPI3*N,75,A!,EVR.EVI,SR,SI,INDI
C CALCULATE S INVERSE AND A INVERSE
CALL ZEROHISIR,75,751
CALL 7rROHLS1!.75.751
CALL HATCPV(AP,AI,76,75,7S,76,76.751
00 IBB I•I,N*3
1.. SIR(!,! 1•1.0.
CALL CINVSEISR,SI,SIR,Sli,75,3*Nl
CALL 8HDINVIAl,3*N,ITESTl
C CALCULATE EMP!LAHIDA*DTI
z•• DO Zlll l•J,3*N
CCIII•OEXPIEVRIII*DTI*DCOSIEV!!II*DTI
Zll QIII•OEXPIEVRIII*DTl*DSINIEVI!Il*DTl
DO z•3 1•1.3*N
DO 2B3 J•1,3*N
lSTlt•ll. D•
STI•I.OB
DO zn K•I,3*N
sn••· o•
ST2•B.DI
CALL CHULTICC!Kl,QIKl,SlRIK,Jl,SIIIK,JI,STI,STZl
2•2 CALL CMUlTISR!I,Kl,Sl!I,Kl,STI,STZ,STR,STil
z•J EX!I,Jl•STK
C CALCULATE ADDITIONAL CONSTANTS
DO z•4 1•1,3*N
oo u• J•I.3*N
£T! I.J I•B.Dll
DO z•A K•1,3*N
ALP•EM!K,J I
IFIK .EQ. JIALP•ALP•J,Oll
284 ET!I,Jl•ET!I,Jl+AIII,Kl*ALP
31B DO 3ll2 1•1,3*N
C:C< I l•ll.o•
DO 312 J•1.3*N
3.2 CCIIl•CCIIl+ETII,Jl*QPIJl
UTUU
END

Table A4-5 Continued

 ~ 434-

SUIROUTINE REDUCEIN,INl
C THIS SUBROUTINE ,ERFORMS THE PHYSICAL ~DEL lEDUCTION FOR
C U~E WITH THE ANAlYTIC SOLUTION.
IMPLICIT REAL"8 (A-H,O-Zl
DIMENSION Tli56,Sil,T2151,51l
COMMON /LINEA/ AAI75,7Sl,88175,8l,DDI7Sl,U18l
IREOLN/ AC175,7Sl,AP!75.751,QP175l,Cll7&,7&l,CZI71,11,
ACI!75,75l
c:
IFIIN .EO. 21 G:O TO 216
HZ•Z•H
N3•3*N
N••4•N
00 liB l•l,N
00 1 (fB ._1• 1 , H
ACI!,Jl•AAIN3+1,N3+Jl
AC!l,N•Jl•AA(N3+l,N4+Jl
ACIN+I,Jl•AAIN4•l,N3+Jl
ACI N•l ,N+J l•AAI N4+1 ,U+J)
T! I I , J l •AA I I, N3+J l
Tlli,N+Jl•AAII,N4+J)
Tl ( N+l.J )•M( N+l,Nl+J)
Tl(N+l.N+J)•AA<N•I.N4+J)
Tl!NZ•I,Jl•AAINZ+I,Nl+Jl
Tl ( NZ+ I .N+J) •A.A( N2+1 ,N4+J >
T2 I I , J ) •AA I N3 +I , J l
TZ! l,N+J l•AA!Nl+l ,N+Jl
T211.N2+Jl•AAIN3+1,N2+Jl
TZCN~t.J)•AA("•~J.J)
TZIN+l ,N+Jl•AAIN4+1 ,N+Jl
lll T21N•l,NZ+Jl•AAIN4+l,NZ+Jl
CALL MATCPVITZ,Cl,S1,56.75,71,N2,N3l
CALL BNOINVIAC,NZ,lTESTl
CALL MATCPV!AC,ACI,7S,75,75,75,NZ,NZl
CALL MATHULT!AC,C1,75,75,7S,7S,N2,N3,N2l
CALL MATMULTITl,AC,56.56,75,75,NJ,NZ,N2l
CALL MATMULTIAC,T2,75,75,51,51,N3,N3,N2l
DO 1*1 l•l,Nl
00 l.fl J•l,N3
Ill APII,Jl•AA!I,Jl·TZ!I,Jl
Zll DO liZ l•l.N
TZI 1,1 l•DDIN3+1 l
TZIN+l,ll•DDIN4•ll
Tl I l.ll•DD!I l
Tl IN+ I , 1 l •DO ( N+l l
T!INZ•I,ll•ODINZ+Il
DO Hl2 ,J•t ,9
TZI 1,1 l•TZI 1.1 1+1111 MJ+l ,JI*U(J l
T21N•l,ll•T21N+I,Il•BBIM4+I,Jl*UIJI
T! I l • I l • T 1 I I ,! l +88 I I , J l * U I J )
Tl IN+!, I l•Tl IN• I, I l+BBI N•l ,Jl*UIJl
t•z T!IMZ+I,li•Tl!NZ•I,ll+IBINZ+I,Jl*U!Jl
CALL MA~CPV<T2.C2,S.,, •• 7&,1,N2,1)
CALL MATMULTIACI,CZ,75,75,75.l.NZ.I,MZl
CAL~ MATHULTIAC,TZ,75,75,$1,51,N3,l,MZl

DO 113 1•1. N3
113 QP!Il•T! I I. I l-TZII, I l
RETURN
£NO

SUBROUTINE OUTRAIV,Ml
IMPliCIT REAL*B IA-H,O-Zl
DIMENSION Y!111l
* COMMON t•EDLN/ :~li~s~~~jAP!?I,?Il,OPI?Il,CII7$,1Sl,CZI7S,Il,

N2•2*N
M3•3*N
NA• .. •N
e?z!!:,!:~i~t.ll
. VII+M41•-C21M+I,II
00 1•1 J•l ,N3
YIN3+1l•YIN3•11-CIII,Jl*V!Jl
11• YIN4+1l•YIM4+1l•CIIN+I,Jl*V!Jl
RETU~N
ENO

Table A4-5 Continued

 -435-

INPUT DATA•

REACTOR rARAMETERS CATALYST rARAMETERS THERMAl VEll rARAMETERS

EPS • 1.5711£+11 CPS • 1.2511£+11 CPT l.lZII£+11
L 1.3BBBE+I2 PS I. 1141£+11 PT 1.8JJ21E+II
Rl 1.1581£+11 TC • 1.5111£•13 KT 1.3911E-II
R1 1.1194£+11

HEAT TRANSFER PARAMETERS REACTOR rARAMETERS

OHS<: • 1.17.. 21::+12 DHIA • -I.U.U£+11 KPIA • -1.29.UE+I2
OHTS • 1.1795£-11 DHI8 • -1.4835£+15 KP18 • 1.2634£•15
OHT<: •
Bl:S
BSG
BTS
1.143&£+11
I.&ISI£+S3
1.1319E+S2
S.7Jfi3E+Bl
DH2A
DH2B
Kl
KIP
..
• -1.2441£+11
• 1.1176£+15
1.1111£+12
1.7524£-11
KP2A
KP28
K2
K3
• 1.4385£+11
• -1.4615£+14
I .1471£+11
I. 1348£+11
8TG
PTZ
6.1 ZIS.IJ£•'1
I.ZSSBE+BI
EAl
EAZ
• .iJ,ti!lb.8'E+f4
ll.l89BE+IS
PTR I.&BII£+11

PROGRAM CONDITIONS•
TIME STEP • 1.11151
MAXIMUM TIME • 5.1111
DISTURBANCE LENCTH • 1119191.1111
RADIAL COLLOCATION POINT • 1.5111
NO. AXIAL COLLOC. POINTS 8
ACCURACY OF CONVERGENCE • 1.111£-18
INPUT FLAG • I
NSES FLAG 1
MAX VALUE OF DT • 1.1861

STEADY STATE CONDITIONS•

XCH4 • 1.&111£+11 Tl • 1.5131£+13

XCO I. 611111£-111 TW • 1.57311£+13
XC02 • 1.15111£-.8'1 PTI • 1.1.8'111£+12
XH2 1.19JJll'£+ll'll' PTI l.lll'll'll'£+11
XH20 • 1.261.8'£-11 MG 1.1593£+12
UG.IJ 1.75118£+12 CPG • 1.6572£+11

AXIAL COLLOCATION POINTS•

I.IBIIE~III 1.1986£-11 1.1117£+11 I.Z312E+II
1.4183£+11 1.5917£+11 1.7628£+11 1.8983£+11
1.9811£+11 1.1111£+11

Table A4-6
Sample Output From Program NLN1IOD
 ~ 436-

INITIAL ;uESSESt

IOAS TEMPERATURE SOLID TEMPERATURE

R • 11.11 II. • II.R R • Itt II. • 11.11 II. • 11.11. It • RI
1 .IIBII/111 1 • 1111111111 1.11Bflllll t.llllfi!IB 1 .flflllflll l.llllllflll
I ,1!1161111 1 .1!1!111111 1 .1111111111 1 • 1111111JI1 1 .IIIIIJIIB 1 ,/IBIIBB
1 .1111/IIIB I .BBIIS/1 I.IIIIBBII 1 ,11/IS/111 1 .11/111111 1.11118/1
1.8/IBIIB 1 .BBiiiiiB 1 .lllil/61116 1 .lllllillillil 1.1illillillil/6 1 .!illil161111
1 .lilllli!lllil 1 .lil/6111111 1 .lllil/11111 1 • 111111Biil 1 .Biil/11111 1.BIIIilfiB
1 .lllillil/111 1 .lllllillilll 1 .1111111111 1 .lllllllilll 1 .fllllilll/1 t .llflllllll
1 .lllillil/111 I .IISIIIIII 1 .llli1111111 1 . lli1111111 1 .llllli11i11i1 1.fllilllllfl
1 .1111111111 1 • lllilllllll I .lllilllllll I .lllil/11111 I .lllillllilll 1 .1111111111
1 .IIB/11611 1 .1111111111 1 .1111111111 1 .llllli1111il I .lllllllilll 1.111i11i1111
1 ,/1/1111111 I .1111111111 l • Sll/11111 I .111111/IB 1 • Bllllllll l • 811111111
\IELL TEMP. CO CONC. C02 CONC. VELOCITY
I .1111111111 11.11&1611111111 11.11151lflllll 1.1116111111
I .1111111111 11.1161i11Jiilllll II .111511111111 1.1111199
1 .1111161116 16 .116/61i11i11111 11.16151illl116 1.1111127
1 • 1616161111 II.BGII/1/IBII II .Ill 516111111 1.11'24311
I . 1111111611 B • 1161611/IIIB 16 .lll5111116B 1 .114257
1 ./11/1/lfl/1 B.IIG/1111111111111 11./115111111/111 1 .116289
l.!illii1Jli116 IJ.IIGIJIJIJIJII /I./1151JIIIIII I .118258
1 • BIIIJBIJ II.BGIJIJiiiiJII 11.111511/IBII 1 .1198711
1.111JIIBB B.B61JIIBIJII 11.111516BIIIJ 1.111867
1.1111111111 11.1161JIIBIIII 1.11151111111 1.11111

STEADY STATE SOLUTION•

II: AS TEMPERATURE SOLID TEMPERATURE

II. • Rll R • RR R • Rl R • 11.11 II. • RR II.. u
1.16l1'1511 1.BBIIJB t.B.IJ1511 1.n111 1.8'11283 Lnttl
1 •.&"11635 I.JJ/16111 l.BBUZ 1.11f18Z8 I.JJ1156 1 • .8'11U2
1.112313 1 • .8'21116 1./11496 1.112626 1 .S:Z654 1.1111868
1 .114569 1 • .8'4-163 I. 12958 1.1149711 I .1147/16 I.B15.1U
t.llll272 1.857118 I .ll.ui'S§ l.llll7:U I .11&21JIJ I - 1119&2
1.117254 1 .B661JS 1 .114693 1.117731 1 .B7.8'AZ 1.8'2219
1.117361 I ,116697 1.114757 1 .1178'2Z 1 .117/163 1./IZZlB
1.117227 t.IJ6578 1.8'4673 1,117662 I.B69.1J5 1.8'2167
1.8'7811 1.8'6365 1.8'4518 1.117449 1./166611 1.1128'114
1.8'6!191 1 .11634 7 I ..U5.1J5 1.8'7326 1.116377 LBI97A
\IELL TEMP, CO CONC. COZ CONC. VELOCITY
I .11118'1111 8'.116111111111 1.115111111 1.1111111
I .JJJJ5A9 11.8'5!16588 II.IIISIIZII 1 • 111162 I
1.8'2324 II .8'573679 ••• 151151 1.8'2955
I . .f458Z 11.11543335 ll • .f15Z437 l • .IJ6U4
I .1163/19 B • .IJ49H.IJII 8'.8'154779 1 .8'9995
I .8'7288 8'.J'A55J'38' ••• 156721 1.131196
1. 1174.1JB B.IIA!!BSJ' 8'.11159.8'25 1.15287
1.87265 .IJ • .IJ38948/I .IJ • .IJ161245 1.16875
1.1171193 .8'.113711699 .IJ.I1613117 l. 17699
1.8'7.8'75 .8'.11368712 11.11161421 1.17939

INITIAL CONTROL VECTOR•

Til •
XC02•
UM
l.IUII
•• 25118'
1 .• 152
UIOI •
l<H20•
~.
1.3333
.... Tl •
l<H2 •
l<CO • 1.1118'
XCH4• 111.11111

Table A4-6 Continued

 -437-

TIME •
••••••
lOA$ SEC
TEMI'EilATUilE SOLJD TEMI'EilATUilE
It • It· It • RR It • Ill It • It· It • llR It • Ill
1 •• 285lJ 1.62376 1.111623 1.63254 1.62226 1 •••613
1 ••6635 1.lJlJ61lJ 1.lJ.U2 1.lJB828 1 .61156 1.••u2
l.lJ2313 1.1121116 1.111496 1.BZ626 I .lJ2654 1 . . . 868
I .lJ4569 l.lJ4163 1 .62958 l.lJ49711 1.114766 1 •• 1564
t.AJ6272 1 • .8'67ll'8 1.B4.8'55 1.BH34 1.B62Bll' 1 .B1962
I .lJ7254 1.116U5 1.114693 1.117731 I .1171142 1.112219
1. lJ7361 I .B6697 1.lJ4757 1.117922 1.1171163 1.112218
1.117227 I.S6578 1.114673 I .117662 1.1169.8'5 I • .8'2167
!.1171111 1.116365 1.114518 1.117449 I .1166U 1 •• 21184
1 .1171128 1 • .8'6377 1 • .8'4525 1 • .8'73711 1..8'6U7 1.111992
IIELL TEMP. CO CONC. C02 CONC. VELOCITY
1 • .8'349.8' .8' . .8'6. . . . . .8'.1115.8'11. . 1.11BBIIII
1 . . . 549 11 • .8'596589 ••• 1561211 11.98479
1.112324 ll.lJ573679 ••• 151151 1 .11.8'76.8'
l.lJ45BZ ••• 543331i ••• 152437 I .AJ4Z34
1.8'63.8'9 .8'.1149471111 11.11154779 1.117639
I .117288 11.11455.8'3.8' 11.11156721 1.1S7B8
1 • .8'74118 •• 114118511 11 • .8'159.8'25 1. 128.8'6
1 .8'7265 11 • .8'399488' 8'.8'16lJ245 !.lUll
1 • .8'7.8'93 11.8'3711699 • • 111613117 1. 15173
1 • .8'71 23 ••• 368712 .8' • .f16U21 1.15459

TIME • S.U.CI'II SEC

GAS TEMPERATURE SOLID TEMPERATURE
R • R8' R • Rlt R • Ill R • Rll R • RR It • Ill
1.113833 1.113619 l..f26116 t..f3838' l.lJ3696 1..1189
I .113576 1.63837 I • .8'2883 1 •• 311411 1.8'4325 1 •• 1548
1.113913 1.1141161 1 .1131119 I .113439 1 .114346 I .111514
l.AJ4915 t.AJ45B7 t.nzn 1.65112 l.AJ4994 1.61614
1 •• 6281 I .1157211 1.lJ41165 1 .8'6731 1.116199 1 • .8'1961
1.117227 I .B657.8' 1 • .f4666 l • .f7714 1 • .f6999 l • .f22.f2
I. 8'7325 I ,116555 l • .f4724 l.lJ7799 l..f7lJ19 1 • .8'2262
I .1171911 I .116532 1 ..14637 1.117641 I .lJ6855 1.62147
I .1169711 1.116317 l • .fUU 1 • .f7422 1.11661lJ I • .fZII65
t • .llli955 t • .II6Z99 1.lJU66 1 •• 731Z l.lJ63Z!I 1 •• 19115
IIELL TEMP. CO CONC. C02 CONC. VELOCITY
1 .113491 1.161.8'1.8'1 ••• 15.8'1111 1 .11/11/11
1..8'1973 •.• 5953!1.f •• 11151'172 I • .8'1141 2
I .12545 .f.IIS689.f3 ••• 15137.8' 1 • .fl461
I .114625 •.• 536.8'98 .8'.111527911 1 • .8'34119
I • .8'63.8'6 ••• 486592 ••• 155188 1 • .f6394
1.117297 ll.lJUU33 ••• 157174 1.119344
1 • .8'7411.8' ••• 4.8'3.8'22 ••• 159516 1.11454
1.•nu JJ .JJ38JJ!iSB ••• I&JJ755 1.129U
1 •• 7.8'83 ••• 361758 ••• 1618311 I .13779
1.n.a3 ••• 359769 ••• 161945 l.U•I•

Table A4-6 Continued

 -438-

STEADY STATE SOLUT!Oih

!;AS TEMPERATURE SOLID TEMPEII.ATUII.E

II. • 11.1 II. • RR R • Rl II. • RS It • lUI. It • 11.1
1 .13858 I .13651 1 .12631 1.13U9 1 .13746 I.IIU9
I ..IU3.1Jl
1 .8"5655
I • .IJ3991
1 .s51•'J
l.ll'2866
1 .13658
1 •• 4&14
1.16133
1 •• 41116
I .15762
1 .• ls••
1.11836
1 • .9'7361 1 • .9'6714 1.BA764 I. 1789.& I. 17296 I. 1231'7
I .S'8336 I .17595 1 .85398 1.19998 1 .S'B.S4 1.12546
1.88652 1 • .9'7878 1 .15597 1 ,lJ9183 1 • .883.81 1.12618
1 .182.84 I .17463 L85381 1.18693 1. 179.9'4 1.1ZU6
I .11775'J 1 ,IJ71f63 1 .11511 a 1.18216 1 • .173&2 1 •• 23111>
1 • .9'7375 1.86698 1.14145 1.17831 I .16956 1.12171
1 • .9'7341 1.16656 1. 84721 1.117784 I..IJ6693 1.12166
YELL TEMP. CO CONC. C02 CONC. VELOC lTV
1.13491 ••• 61111111 ••• l!;lJ.IJlJlJ t ••••••
1.1.112 8.1595285 1 • .1151176 1.U531
1.85688 8.8567781 1.115H24 1.82493
I .117397 1.153.1768 . f.8153B43 1.1'5451
I .18387 1.1476671 1.1155718 1.18211
1.18692 1.1433679 1.1151911 1.1162&
1.18261 1.1389593 1.1161331 I.IZZ38
1.118.11 1.1367156 l.lllilU& 1.13487
1.17484 1 • .9'348715 1.1162672 1.UI15
1 .74!\' •.• 'UI;771 •. IIU,71t7 I.U:'l:'l'

Table A4-6 Continued

 -439-
SUBROUTINE DIMLES
THIS SUBROUTINE CALCULATES THE DIMENSIONLESS PARAMETERS
IMPLICIT REAL*& <A-H,O-Zl
REAL*& L,KT,MG,K6,KBP,KC20,KP2A,KP2B,K!P,K2P
COMMON /REACP/ EPS.L,RB,Rl
/CATLS/ CPS,PS,TC,OC
/THWEL/ CPT,PT,KT
t;ASPA/ CPG,PTz,PTR,MG,PGS,U;S,VM
/HEATT/ OHSG,OHTS,OHTG,BGS,8SG,8TS,BTG,B~.BVS
/OPCON/ SCHI,SCO,SC02,SH2,SHZO,PTB,PTI,STB,STW
/REACI/ DHIA.OHIB,DH2A.OH2B.KB,KBP
/OIMLF/ ALS.ALG.ALT,BES.BEG.GAS.GAG.GTS,;T~.OEl,
OE2,Sil,SIZ.SI3,PH12,PH13,PH1,PHZ,PH3
/REAC2/ EXPA,EXPB.EXPC,KC2B,KP2A,KP28,KIP,K2P,EAI,EA2
DATA PI /3.1415926S40B/

VT•Pt•Ft.f**Z*L
VB•P1*L*tkl**2·~••*2}
u•L/DC
AR•RI/OC
AO•L/P.I
EMI•I.BO.f-EPS
ALS•TC/tPS*CPS*EMI*L*UQSJ
ALG•I.DB/tPTZ*EPS*azl
ALT•KT/{PT*CPT*L*U~$)
8ES•ALS*l**2/R1**2
BEG•AO/PTRIEPS/AR
GAS•OHSG*l/(VB*PS*CPS*EMl*UGSl
GAG•OHSG*L/\VB*PGS*CPG*EPS*UGSJ
GTS•OHTS*L/\VT*PT*CPT*VGSJ
QTG•OHTG*L/tVT*PT*CPT*UGSl
OEl•~LwDH10•<PTS•SCO>••(£XPA•EXPB>•KeP/CUGS•C~S•ST•>
DEZ•-L•OH2B*SCO*•Z*KB/tUGS*CPS*STBl
SII•MG*PS*EMI*\PTB*SCOl**\EXPA+EXPBl*K.fP*L/(£PS*UGS*PGS*SCOI
SIZ•MG*PS*EMI*SCO*K.f*l/lEPS•u;s•PCOSJ
513•512
PHI2•DHIA*STB/OHIB
PHI3•0H2A*STB/OH28
8\114•BTG
BVS•BTS
PHI•RB/RI
PHZ•PTI /PT.f
PH3•STW/STB
RETURN
END

SUBROUTINE INITIAL (Y,U,IFLAG)

C THIS ROUTINE INITIA,IZES THE SYSTEM
IMPLICIT REAL*B IA-H,O·Z>
REAl*8 MCO
DIMENSION Y!l.f.fi,U181
COMMON /OPCON/ SCH.,SCO,SCOZ,SMZ,SMZO,PTB,PTI,STB,STW
IGASPA/ CPG,PTZ,PTR,MQ,P;S,VGS,UM
/COHAT/ At26,25),.(26,2,),R(Z,),H
DO l.flf 1•2,4
3.fB Ul I l•l.DB
UtI l •STVISU
UISI•SC02/SCO
Ul6l•SH20/SCO
U(7J•SHZ/SCO
U! Bl•SCHHSCO
UM•l. DB
IF!IFLAG .EQ. IJCOO TO 2BB
READ<S.ll TCO,TS,TT,Xl,X2,VEL
DO !BB l•l,N
Yt3•N+l)•XlfSCO
Y! 4*N+I l•X2/SCO
Y!S*N•I>•VEL*PTB!IIPTI·PTBI*R(I+li+PT.f!/UGS
Vt!J•TS/SU
Y!N+IJ•TCO/STll'
Y12*N•Il•TT/STB
1BB CONTINUE
Y!6•N•l>•VEL*PTI/PT1/UGS
Yt6*N+Z>•TS/ST.0'
VI 6*N+3 J•TS/STB
V!6°N+4J•TG/STB
YI6*N+Sl•TIO/STB
Yl6*N•6l•XIISCO
RETURN
2.fB DO 2.fl l•I,N
READ\5,1) V\N+ll,Ylll,YIZ*N+Il.XI.X2,YIS*N+I)
E•tSCO+SC02-Xl-XZJ/II.DB·2.0.f*Xl-2.DB*X2J
D•tSCOZ-X2+2.0.0'*1SCO•xz • Xl*SC02JI/II.O.f•2.0B*XI•Z.DB*X2)
YI3*N•I>•<SCO-E•Dl/SCO
Vt~•N+l)•(SC02-0l/SCO
Ul CONTINUE
READIS,II Vl6*N•21,YI6*N+3l,Y16*M+4l,Y16*N+S),YI6*N+Il
UTURN
FORMAT!9F8.2J
UD

Table A4-7

library Program UJB

 -440-

SuBROUTINE SETUPS IOT.,TMAX.Ol,N,RR,£P,IFLAG,NFLAe,NP,IFI

t THIS SUBROUTINE READS IN THE INPUT DATA AND MAKES PREliMINARY
C CAlCULATIONS.
IMPLICIT REAL•G IA-H,0-%1
REAL•a L,KT,MG.KI,K.P,KP2A,KP28,KCZI,KIP,KZP,K2,K3
COMMON /REACP/ EPS.L,RB,RI
/CATLS/ CPS,PS,TC,DC
/THWEl/ CPT.PT,KT
/~ASPA/ CPG.PTZ.PTR,MG,PGS.UGS,VM
/HEATT/ OHSG,OHTS,OHTG,BGS.BSG.8TS,BTG,8WG,8WS
/OPCON/ SCH4,SC0.3COt,SHZ,SH20.PT6.PTltSTM,STW
IREACI/ DHIA,OHIB,DH2A,DH28.KB.KBP
!REAC2/ EXPA.EXP8,EXPC,KC2B.KP2A.KP28,KIP,KZP,EAI,EAZ
DATA RG.RGP,Pl /82.BS~~OB,I.987DB,3.1~1592654DB/
READIS,II EPS,l,RB.RI
REAO!S,Il CPS,PS,TC,OC
REAOIS,II CPI,Pl,n
READIS,ll PTZ.PTR,UGS
READIS,ll OHSG.OHTS,OHT~,8GS,8S •• BTS,ITG
lEAOIS.Il SCH~,SCO,SC02,$H2,SH20,PT.,PT!,ST.,STW
lEAOIS,Il OHIA,OHI8,0H2A,DH2B,KB,KBP
READIS.Il EXPA,EXPB,EXPC,KP2A,KP2B,K2,K3,£AI,EA2
REAOtS,Z> DT#,THA~.RR.D~oTS,NtHP,N£P,JflAC,JF,HFLAe

CALL CPCALCISH2,SCO,SCOZ,SHZO,SCH4,ST•• MG,CPG,CPG!,CPG21

PGS•MG•PTI/RG/ST•
KIP•KZ•PTB 0 SCO
UP•K3°PTB*SCO
KCU•I.DB
EP•lJJ.D.If••"EP
KBP•KBP•Ts••!-.31
PRINT 3,EPS.CPS,CPT,L.PS,PT,RB.TC,KT,Rl,OHSG,OMIA,EXPA,
OHTS,DH!B.EXPB,OHTG,OHZA,EXPC,BGS.OHZS.KPZA,BSG,KB,
KPZ8.BTS,KBP,KZ,BTG,K3.PTZ,EAI,PTR,EA2
PRINT •• OTB,TMAX,OL,RR.TS,N,EP.IFLAG.NFLAG,OTB•~.OB••ttF-11
PRINT S,SCHA,STB4~CO,S1W,S~UZ,PI.,SHZ,PTJ,SHZO,"W,UW$,C,G

OTJJ•OTB•VGS/L
TMAX•TMAX*UGS/L
DL•OL•VGS/L
RR•ll.OB-RB/Rli*RR+R./RI
EAI•EAI/RGP
EAZ•EA21RGP
RETURN
I FORMATI908.21
2 FORMATISF8.2,/618l
3 FORMATllHI,////,3X,'INPVT DATAr',///9X,'REACTOR PARAMETERS',eX,
'CATALYST PARAMETERS',6X.'THERMAL WELL PARAMETERS',!/
9X, 'EPS •' ,EIZ.~.9X. 'CPS •' ,EI2.~.9X,'CPT •' ,£12.4/
9X,'L •',El2.4,9X,'PS •',EIZ.4,9X,'PT •',E12.4,/
9X,'RB •',E12.4,9X,'TC •',EI2.4,9X,'KT •',EIZ.4/
9X, 'RI •',EIZ.H////6X,
'HEAT TRANSFER PARAMETERS',I9X, 'REACTOR PARAMETERS'//
9X, 'OHSG •' ,EIZ.~.9X, 'OHIA •' ,EI2.~,9X, 'EXPA •' ,£12.4,/
'X, 'OHTS •' ,£:12 • .& ,IX, "OWl I •' ,£12. 4. 9)(, '£)(PJJ: • ' , £12. of. .I
9X, 'OHTG •',EI2.4.9X,'OH2A •',EI2.4,9X, 'EXPC •' ,E!2.~,/
9X, '&GS •' .EIZ. 4, 9X, 'OHZB •' ,EI2. 4, 9X, 'KP2A •' ,EIZ. 4,/
9X, 'BSG •',EIZ.~.9X,'K.f •',El2.4,9X, 'KPZB •',El2.~,/
9X,'8TS •',E1Z.4,9X,'k6P •',El2.4,9X,'K2 •',E12 .• ,/
9X,'BTG •',El2.4,36X,'K3 •',Et2 •• /
9X,'PTZ •',El2.~,36X,'EAI •',ElZ.V
!x,•PTR •".E1Z .• t36X-"EAZ •".E1Z •• )
4 fORMATl////,3X,'PROGRAM CONOITIONS•'//,9X.
'TIME STEP •',F8.4,/9X,'MAXIMUM TIME •',F8.4,/9X,
'DISTURBANCE LENGTH •',Fl5.4,19X,
'RADIAL COLLOCATION POINT •',F8.4,/9X,
'START TIME lHRSl •',FIS.4,/9X,
'NO. AXIAL COLLOC. POINTS •',I~,/9X,
'ACCURACY OF CONVERGENCE •',ElB.J,/~X,'INPUT FLA~ •',14
/9X,'NSES FLAG •',!4,/9X,'MAX VALUE OF DT •',F8.4)
S FORMATI////,3X,'STEADY STATE CONDITIONSo',///,9X,'XCH4 •',
El2. 4, 9X, 'T.f •' ,E!Z.4, /9X, 'XCO •' ,El2. 4, 9X, 'TW •',
E12.4,/9X,'XC02 •',E12.4,9X,'PTB •',EI2.4./9X,'XH2 •',
El2.4,9X, 'PTl •' .EJ2: • .4,/9X, 'XHZO •',El2:.A,9X, 'HG •'
£12.4,/!X,'UG. •'.ElZ.4,,X,'CP; •',EIZ.4l
uo

Table A4-7 Continued

 -441-

SUIROUTINE OUTPUT IY,U,T,L,u;S,PHI,""l

C THIS ROUTINE OUTPUTS THE CALCULATED RESULTS.

IMPLICIT REAL"I IA-H,O-ZI

1lE,t,L•8 l
DIMENSION Yltffl,UIII
COMMON /COMAT/ AIZS,251,8125,25l,RtZSI.N
/OPCON/ SCHA,SCO,SC02,SHZ,SHZO,PTIJ,PTI,STIJ,S~
/RADIAl \It llJl ,\/PI 9,2 I ,OETAt 2 l ,RR
DATA RG /82.f544Df/
lr<MM ,£Q, ll P~INT I
IFtMM .EQ. 21 PRINT 6
IFtMM .EQ. 31 PRINT 8
PHZ•PTI /PTf
N2•2*N
M3•3*N
N••••N
"5•5"""
M6•6*N
NP2•N+2
TA•T*L/UIOS
IFtMM .EQ. fiPRINT 7,TA
CALL ENOPTS ITTIJ,VIlJ,YlNPl,YZf,YZNPI,TTNP!,Y,UGfl
PRINT Z
PR !NT 3, T1UYI N6+Al, TTIJ, Z, PHIl, Yt N6+Al, TR tYt N&+A I, TTf, 2 .1. DlJl,
TR tYt N6+ZI, TTIJ, !,PHIl, Vt N6+2 I, TRt YtN6+2 l, TTf,!, t .Of>
DO If! I•I,N
PRINT 3, TR tVt N+l l, Vt NZ+I l .z ,PHIl, Vt N+l), TRIYt N+l l, Vt N2+1 I ,2,
I.DlJl .TRtYt ll,VtNZ•l ),!,PHIl ,Vt I l ,TRtYt I l ,YtNZ+I l ,I, l.Dfl
PRINT 3,TR!YtN6+5l,TTNPI.Z.PHll,Y!N6+51,TRtVtN6+5l,TTNP!,Z,I.DlJI
,TR!VtN6+3l,TTNP!,I,PHli,YtN6+3l,TRtYtN6+31,TTNP!,I,I.OlJI
PRINT
PRINT 5, TTf,Yif"SCO,Y2f"SCO,UGB
00 IIJZ l•I,N
T~l•l Og-,_n«•c~~O*tU(£)-V{N3+l)+U(Sl-V(N,+1)))
PRINT 5,Y!NZ+li,YIN3+11*SCO/THI.YIN4+11•SCO/THI,YIN5+11
THl•I.OlJ·2.0lJ*lSCO•IUI41·YINPI+Ut51·YZNPlll
PRINT S,TTNPI,YINPt•SCO/THI,VZNPI"SCO/THI,YtN6+tl
RETURN '
FORHATIIHI,//ZX,'INITIAL GUESSES• ' , I l l
FORHAT!I&X,'QAS TiMPiRATURE',2SX,
'SOLID TEMPERATURE',//8X,'R • R.',7X,'R • RR',7X,
'R • RI ' , 9X, 'R • 11.6' , 7X, 'R • RR' , 7X, 'II. • II. I ' I
FORM,t,TISX,3tF9.5,4Xl,2X,31F9.5,4Xll
FORMAT<tl9X.'IIElL TEMP.',SX,'CO CONC,',5X,'C02 CONC.',AX,
'VELOCITY' l l
s FORMAT( l9X. F9.5 ,SX ,Ftf. 7 ,3X ,F 16.7, 3X ,F9.5,3X ,F9,S I
•
7
FORHAT(//Z><,•srtADV STATE: SOLVTION1
FORMAT!//,2X,'TIME •',fll.4,' SEC'/)
'•//)

I FORMAT!//,2X,'ACTUAL STEADY STATE SOLUTION• ',//1

END

SU8ROUTI"E ~ADIAL

C THIS SUBROUTINE CALCULATES THE CONSTANTS FOR THE RADIALLY LUMPED

C MODEL.
IMPLICIT REAL•& lA-H,O-Zl
COMMON /DIMLE/ ALS,AL,,ALT,BES.BEG,~AS,GA,,GTS,GT!O,D£1,
OEZ,Sll,SlZ,SI3,PHlZ,PHI3,PHl,PHZ.PH3
/II.AOIAI \It 11Jl,IIP<9,2l ,DETAI21 .RR
/HEATT/ OHSG,OHTS,OHT;,BGS,BSG,BTS,8T;.8WG,8\IS
vc,••.o•
\It IBl•lJ.DB
l•l
S!•BTS
SZ•B\IS
S3•tOAS
S.•BES
SS•!OTS
DtTAlll•lPMI•PH1°*2*lRR-l.D.l+RR••z•ti.O··PHII·II.II.)•SI 0 $2
+ tZ.DlJ*{PHI·RRl+RR*"2-PHl*"21•St
+ 1-I.DIJ•RR""2•2.Df"IPHI·PHI*RR1l*S2+Z.Of•PHI·Z.Df

liP l t, I I• t PHI*Sl·l. D•l • I S2+2 .OBI •($2+1. 0f)•IPHI .. 2*S 1-2. o••PMI I
11Pt2,ll•St•tS2•1.Dfi"RR**2·SI"RR*IS2+2.Dfl
IIP13,ll•tS2*RR*IPHI*"2"S1·2.D••PHil•S2*1PHl*SI·I.OlJl•RR""2)
WP(4,I••S2•(PHt**2*S1-2.D.*PHl>•Sl*(S2•2.0.)
IIPIS,ll•St•tS2•2.0Sl-St•SZ*RR••z
IIP16,1l•tSI"S2•RR*"2-S2"1PHl**2"SI•2.06"PHill
IIP17,11•Sl•ISZ+I.D·l-SZ*IPHI"Sl-I.Ofl
11Pt8.1l•RR"SI*$2•SI*IS2+l.DBl
\IPI9,11•l$2•tPHI*SI·I.DBl·RR•SI*SZl
Ill 1•1"3-3 l• 14. OIJ*IIPI 7, 1 l+IIP 14, l l /RR l"U/OETAl l l-S3
W( l-tl '*3•3 )•\., OB'"''w'P (8 • 1 }+\JP ( '$ ~ t > /AR )•$<4/0ETA( l l
lll3+1"3·3l•t4.01J"IIP!9,!1+\IPI6.1l/RRI*S4/DETACII
II( Hl·l ) •SS• I liP ( I , I l +liP! 4, l l •P HI +liP 17, I l •p Htnz l /DETAt I I
111,•111 9l+S5*!\IPI 2,1 )+1/PIS,I l*PHl•IIPl8,1 l*PHI"*21/DETAI I l-SS
II ( llJ l •Ill IJ!'l +SS• I liP l 3, I l +1/P t & , I I *PH 1•1/P! 9, I I *PH I ""2 liD ETA( I l
IF!I .EO. ZIRETURN
1•2
S I•BTG
SZ•BIIG
S3•GAG
S4•8EG
SS•GTG
;o TO lfB
ENO

Table A4--7 Continued

 -442-

SUBROUTINE rMI !Y,Vl

C THIS SUBROUTINE DEFINES THE ALGEIRAIC EQUATIONS FOR USE IV NSESl
I"PL!CIT REAL*B !A-H,O-ll
DIMENSION Y!l.&'.fi,V<IJ
COMMON /COMAT/ A!ZS,25J,II2S,ZSI,~I251,N
/OIMLE/ ALS,AL;.ALT,BES.BEG.GAS,GAG,;TS,GT;,DE!,
OEZ.Sil.SIZ.Sll,PHIZ,PH!3,PHI,PH2,PH3
/RADIAl IJ!l.lll ,lo'PI9, 2 l ,OETA!2 l .RR
/HEATTI OHS;.OHTS.OHT;,BGS.BSG,BTS.&T;,B~,8VS
"'2•N+2
N2•N•2
trt3•N•3
fUi•H*4
N5•N*5
"'"""*'
TAU•PHZ·!.O.f
CALL EN OPTS !TTl, VII, YlH,I, YU, V2NP I, TnPI, Y,UUI
00 I n I•l.H
l¥1•1+1
~T•TAU•~tlPl>•l.,l

CALL REAC tYIN3•1J,VIN4+!1,V!IJ,YIN+IJ,PT,RIP,R2PJ

Sl•SUM<V,.f,8,VIN6+ZI,V!H6+3J,I,HI
SZ•SUM(Y,N,A,Y!N6•4l,YIN6•5J,l,NI
SJ•SUM<Y,N,B,VIN&•41,V!H6+5J,I,NJ
S4•SUM!Y,H3,A,VI.f,VINPI,I,Hl
S&•SUM!V,N4,A,YZI,VZHPJ,l,Nl
S8•SUM! V,N2 ,B,l.IO.f, TTNP l,I,Nl
VCJ> • ALS*Sl • Wlll*V<JJ • W(2l*Y<N2+J) + GAS*Y(N+l)
+ DEI*RJP*II.O.f+PH!Z*Y!IJJ + OEZ*RZ,*II.O.f•PH13*Yilll
+ W(3l*PK3
VCJ+N) • -VtNS+I)*$2 • tALQ*$3 + W(.}*Y<N•I> + WtS)*YlNZ+J)
+ GAG*Y!IJ + \1<6l*PH31/PT*YIN•IJ
V41+H2l ALT*U + 'olt7l*YIIl + IIC81"Y!N•ll + 1119l*YIN2+1J
\/( lii*PH3
V!I•N3l • -V!NS•Il*S4 + SIZ*V(H•Il*kZPIPT- SII*V!N+IJ*k!P/PT
Vli+N4l • •YINS+IJ*S6 - SI3*Y!N•I J*R2PIPT
ll.f CONTINUE
V!H6+2l•SUMIV,.f,A.YIN6•2l,YIN6+3),.f,Nl+IGS*!VIN6+•1-YIN6+21)
V!N6+3J•SUMIY,.f,A,YIN6•2l,YIN6•3l,N•l,Nl+8;S*!YIN6+3J•YIN6+1JI
VIN6+4l•SU~!Y,N,A.Y!N6•4l,VIN6•SJ,I,Nl+BS'*(Y!N6•21-YIN6+4l)
•ti.O.f-J.OI/YlN6+4Jl/Al~
V!N6•5l•SUM!Y,N,A.YIN6+(J,VIN6•5J,N+I,Nl+ISG*IYIN6+51·YIN6+3ll
pt.•ph2
CAll REACIYINPI,YZNPI,YIN6+31,Y!N6+5l,PT,RIP,RZPI
VIH6•6J••VIN6•ll*SUM(V,N3,A,YII,YINPI,N+I,NI
+SI2*V!N6+SI*RZP/PT - Sll*Y!H6•SI*kiP/'T
DO Ill K•I.N
ll.f VINS+Kl•YIN+KJ*SU"IY,H5,A,UG.f,VIN6+1J,K,NJ•YIH5+KI*SUM!Y,N,A,
Y("6+,),Y("6~,t,K,~~~Y!H5•K>•Yt~•K>•TAU/fTAU~RfK•ll+l.0.)
V!H6+ll•VIN6•51*SUMlV,N5,A,UGI,Y!N6+1 >.~•I,Nl•V!N&+IJ"SU~!Y,N,A
, VI N6•4 J. YIH6+51. N+ I,N l•Y( N6•1 I*YIH6•5 J*TAU/ (TAU+I,O.f)
RETURN
END

SUBROUTINE FNZIKK,Y,VI
C THIS SUB~OUTINE DEFINES THE AL;EBRAIC EQUATIONS FOR USE BY NSES2
IMPLICIT ~EAL"8 !A·H,O-Zl
DIMENSION Y!ll.fl
COMMON /AlGEB/ Al!6.fl,A214l
/COMAT/ Al25.25l,BIZ5,25>,R<25J,N,I8
/DIMlE/ ALS,AlG,ALT,BES.BEG.~AS,,A,,GTS,GT;,OEI,
DE2.Sll.SlZ,Sl3,PHIZ.PH13,PHI,PH2,PHJ
/RADIAl \l!l.fJ.\/Pt9,Zl,OETAIZI,RR
/HEATTI UH~~,OHTS,OHTG.IGS,I$G,ITS.ITQ.8WG,8W$

NP2•H+2
N2•N*Z
trt3•N*3
"4•N*'
NS•S•N
fii6•N*6
TAU•PH2-l.DI
CALL EHOPTS tTTI,Ylf,VINPI,YZI,YZN'l·TT"Pl,Y,UG.fl
V•/6,01
IF IKK .GT. N51~0 TO Z.f.f
If IKK .;T. IUJ>;O TO 111
I ' !KK .~T. N3lCO Tn 1211
IF IKK .GT. N2l~O TO 13/6
IF !KK .GT. N>;o TO IU
I•KK
I Pl•l+l
PT•TAU*RI!Pll+l,O.f
CALL REAC tVtN3~t).YtN,+l).V(J).VCM•l),PT.~1P.l2'>
SI•SUMIY,I.B,V!N6•21,VIH6•31.1,NI
V• AL.S•Sl • Wll>*Vtl) + \/t2J*Y<N2+1) • (i;AS*VOf+J)
• + OE!*RIP*II.D.f+PHIZ*Yilll + OEZ*RZP*II.O.f+PH!3*YIIJJ
tt + Wt3)*ftH3
RETURN

Table A4-7 Continued

 -443-

!U I•KK•N
IPI•l+l
PT•TAU•R!IP!l+!.Df
S2•SUMIY,N,A,YIN6+4l,YIN6+5l,I,Nl
S3•SUMIV,N,B,YIN6+4l,YIN6+5l,I.Nl
V• -YINS+ll*SZ + IALG*SJ + Wl4l*YIN+Il + W15l"YINZ+Il
+ ;A,*Yill + Wl6l"PHJI/PT"YIN+Il
I F'TUIN

131 I •KK•N2
S8•SUMtY ,N2 .B.I.fOI, TTNPI, I, Nl
V• ALT*SB + W171*YIIl + W181*YIN+Il + W19l"YIN2+ll
+ WI lfl*PHJ
RETURN
121 I•KK•N3
IPI•l+l
PT•TAU*RIIPil+l.Of
CALL REAC IVIN3•Jl,VIN4+Jl,YIIl,YIN+Il,PT,RIP,RZPl
S4•SUMIY,NI,A,Ylf,YINP!.I,Nl
V• •YIN5+ll"S4 + SIZ*YIN+!J•R2P/PT • SII*YIN+ll*RIP/PT
llETUilN

lllJ l•KK•U
IPI•I+l
PT•TAU*RIIP!l•l.OlJ
CALL REAC IYIN3+ll,YIN4+1l,Yill,YIN+!l,,T,RIP,l2Pl
S6•SUMIV,N4,A,Y2f,VZNP!,l,Nl
RETURN
ZflJ K•KK•IIS
KPI•K+l
!FIK .LE. N+ll GO TO Iff
L•K·N·l
IF IL .EQ. ll V•SUMIY.f.A.YIN6+Zl.YIN6+3),f,Nl+B~S*IYIII6+4l
-YIN6+2ll
IF IL .Ea. Zl V•SUMIY,f,A,VIN6•Zl,YIN6+3l,N+l,Nl+B'S"IYIII6+3l
-VIN6+Sll
IF (L .Ea. 3l V•SUMIY,N,A,VIN6•4l,VIN6+Sl,lJ,Nl+BSG*IYIN6+Zl
·YIN6+411-Il.Df-I.Df/VIN6+4ll/AL;
IF IL .Ea. 4l V•SUMIV,N,A,VIN6+4l,VIN6+SI,N+l,Nl+BS;•!YIN6+5l
-VlN6+3))
RETURN
Iff IF IK .Ea.
N+ll V•YINS+Kl"VIN6+5l*TAU/ITAU+l.Dfl+VIN6+SJ•
SUMIY,NS,A,UGf,VIN6+1l,K,Nl·YINS+Kl*
SUMIV,N,A,VIN6+4l,YIN6+5l,K,Nl
IF IK .NE. N•l l V•YINS•Kl*YIN•KI*TAU/IRIKPI l*TAU+l.Dfl+YIN+KI•
SUH(V,NS,A,UQS,V<HG•1J,K,H)-V(H5+K)*
SUMIY,N,A,YIN6+4l,YIN6+Sl,K,Nl
RETURN
END

SUBROUTINE FN3 IY,Vl

IMPLICIT REAL*S IA-H,O-Zl
DIMENSION Vllf.IP,VCil
COMMON /LINEA/ AAI7S,75l,BBI7S,Bl,ODI75l,UI8l
/LAL<:SI SC 4,4 l ,HI Zf ,Ul ,R l IZI,4 l ,ALI 7l
/COMAT/ ACZS,ZSl,BIZS,ZSl,RIZSl,N
DO 18. l•l.N•ti
VI I 1•00! I l
DO Ill J•l,N"5
Ill Vlll•VIIl+AAII,Jl•Y!Jl
00 lfZ J•l,S
112 Vll>•VIll+BBII.Jl•U!Jl
1•1 CONTINUE
RETURN
END

Table A4-7 Continued

 -444-

~URROUTIN£ LREAC!YS.RLI.RLZI
C THIS $UBROUTIN£ CALCULATES TME COEFF!C!EWTS FOR TNt
C LINEARIZED RATES.
IMPLICIT REAL*8 CA•H,O-ll
REAL*8 KPZA,KPZB,KCZB.KIP,K2P,KPZ
OlMENStO,. Dltftr;,ql,tl'1'tt§.q).VStl••>
COMMON /OPCON/ SCH~.SCO.SCOZ,SHZ.SHZO,PTJ.PTI.STJ,STV
/~EACZ/ EXPA.EXPB,EXPC.~CZI.KP2A,KP28,KIP,KZP,EAI,EAZ
/COMAT/ AC25,25l,8125,2SI,RC251.N
/LINEAl AAI 75,75 I .89175,8 I ,DOC 75 I ,Ul 81
DATA fi,FZ /J.8301,J.I70B/
DO 1611 f .. l,K
YI•YSI3°N+ll
V2•VS14"N+Il
TS•YSC I)
T'•YSIN+I I
PT•CPTJ/PTJ•I.OBI*R!l+\l+\,DJ
"ti•$i.•TS•T$
KPZ•OEXPCKP2A+KP2B/T$/STJI
THI•SCO*IUI41+UCSI-YI•Y21
THZ•SCO*CUCSI-VZI
TH3•1.01-2.0B*THI
VH2•1SHZ-J.OI*THI-TMZI/SCO
VHZO•ISHZO•THI•THZI/SCO
YCH••!SCH4+THll/SCO
PA•Pi•TH3
AAA•l.DI+KIP*PT"Yl+KZP*PT*YHZ
RLII!,II•OEXP!-EAI/STJ/TSI"PT**CEXPA+EXPBI*YHZ**EXPS*Yl**EXPA
tAAA•*EXPC
RLlll.Zl•RLICI.Il*EAI/TT
RL1ti.3)•RLlCJ~l)*(J•EXPB/YHZ+EXPAtYl-Ex'c•,r•{~l,+J•~zPJrAAA>
RLICI,~I•RLICI,II*C4*EXPB/VH2•4*EXPC*K2P*PTIAAAl
RLJ!I,Sl•RLIII,Il*C-l*EXP8/VHZ+3*EXPC•~zP•PT/AAAl
RLJII,6l•RLI!l,li*C·A*EXPB/VH2•A*EXPC*KZP*PT/AAAI
RLI I I. 7l•B. DB
RLICI,Bl•RLICI,II*CEXP8/YH2·£XPC*K2P*PT/AAAl
R lt ( t • g) · • . o•
AM•F 1 +f:'2•PA*PT.f
CC•AAA*OEXP I -EAZISU/TS I!TH3/TH3
DOD•J.og•vz•~CZB*Yl/KPZ
£E•4.og•VZ+Z.OI*KC21*Vl/KP2
RL2CI,li•IYI*YHI-KCZI/KPZ*VI*YH201*CC
8BB•\2.Dg•rz•~TM•PiJAAA•4.DS!TM,>*RL2(!.l)*~CO
RLZII,Zl•RLZII,Il*EA2/TT-KC2B*Yl*VH20*KPZ8*CC/KP2/TT
RL211,3l•BBB•CC*IDOO·KCZI*YHZO/KPZl
RLZII,Al•BBB•CC*CEE+VH2l
RlZII,Sl••BBB-CC*OOO
RLZII,6l•·BBB·CC*EE
Rl211,71•·CC*KCZB*YI/KP2
RLZ(l.I>•YZ•tC
Rl211,91•1.101
Ill CONTINUE
II.ETUIIN
EMO

Table A4~7 Continued

 -445-

IUI~OUTINE .SETU,!Y,lLI,lL%1
C TMIS SUBROUTINE SETS UP THE STATE MATRIX 'A', CONTROL MATRIX 'I' AND 'D'.
IMPLICIT REAL*8 IA-H,O-ZI
llEAL•e MG
DIMENSION VI l.l'.e'l ,RL 11 15,9 l ,RLZI l S, 9), F 121, Z.l'l ,QIC, 2.1', 2.1') ,P!Ul
COMMON /LINEA/ AAI75,75l,BBI75,8l,00!7Sl,U!Bl
/LAL;S/ SI4,41.HI2.1',4.1'1,RI!Z.I',41,AL!71
/OIHLE/ ALS,AL~.ALT,BES,BEG,GAS,;A~.GTS,GTG,OEl,
DEZ,Sil.SIZ,SJ3,PHIZ,PHI3,PH1,PHZ,PH3
/RADIAl Ill llJI,\oiP!9,2l,DETAIZI,RR
/COMA'f/ AIZ~.Z~I,81Z5,Z51,RIZ5l,N
/GASPA/ CPG,PTZ,PTR.HG,PGS,UGS,VH
/OPCON/ SCH4.SCO,SC02,SHZ,SHZO,PTIJ,PTl,STIJ,STII
/HEATT/ OHSG,OHTS,OHTG,BGS,8SG,8TS,8TG,811;,8\IS
NPI•N+I
NPZ•N+Z
HZ•H*Z
N3•N*3
N4•N*4
NS•N•S
H6•N*6
TAU•PHZ·I.DIJ
SCOZB•SC021SCO
SCHH•SCH4/SCO
SHZB•SH2/SCO
SHZOB•SHZO/SCO
CALL CPCALCISHZ,SCO,SCOZ,SHZO,SCH4,STJ,M;,CPG,CPGI,CPGZI
C CALCULATE THE S MATRIX
Ml 1,1 I•BGS-AI 1,1 l
MII,ZI•·Ail,NPZI
AAI !,3l•·8GS
AAil,4l•lJ.DlJ
AAIZ,Il••AIHPZ,Il
AAIZ,ZI•-AINPZ,HP21·BGS
MCZ,3)•1.DI
MIZ,4 l•BGS
AAIJ,Il•·BSG
AA13,2l•.I'.DIJ
AA(3,3l•I.DIJ/ALG/YIN6+41""2+8SG•AII,Il
AAI3,4l•·At I,NPZl
AA!4,1 l•.I'.D.I'
AAI4,ZI•DSG
AA14,3l•·A!NP2,ll
AAI4,4l•·AINP2,NP2l-8SG
CALL BNOINV IAA,4,1TESTl
CALL HATCPV IAA,S,75,75,4,4,4,4l
C CALCULATE THE F AND H MATRICES, ALONG WITH MATRIX Q
DO 2.1'1J l•l,NPI
Ul PI I l•TAU"RI 1+1 l+t,l)•
ALL•TAU/P!NPll"YIN6+1l+SUH<V,NS,A,UGIJ,Y!N6+1l,NPl,Nl
DO lflJ J•I,N
JPI•J+l
=~==:~::::::}:~=:::=:~::::::~:::3=::
S3•SI3,3l 0 AII,JPil+S(3.4)•A(NPZ,JPll
S~•SI4,3l•A<l,JPI)+SI~,Il*AINP2,JP1l
DO 1.1'1 l•l,N
lPI•I•l
HII.Jl•ISI*AIIPI.ll+SZ*A!J,I.NP2ll•VtNS+Il
HI!, N+J l •I AI IP I,JP I l+S3*AI IP I ,I l+U 0 AI lP 1 ,NP2 l l*YINS+l!
Ql I,I,Jl•ALG/PI I l 0 III!PI,l l*Sl+IIIP1,NP2>*S2l
QIZ,l,Jl•·AI !P1,1 l*Sl•A( lPI,NPZl 0 SZ
Ql 3, l , J l •ALI:/P I I l • I 8 I I PI ,JP I l •B I I PI, I l 0 $3+8 I l PI, NPZ l *S4l
Ql 4, l , J l •·All PI •JP 1 l ·AI I P1 , I l *S3·A< I PI , HPZ l *$4
Ill AAil.Jl•YIN+ll*AIIPI,JPil
4AtMPt.J>•VfN6+S)•A(NP2.J,l)
AAIJ,Jl•AAIJ,Jl+TAU/P!Jl*YIN+Jl•SUHIY,N,A,VIN6+4l,YIN6+5l,J,Nl
HIJ, N+J l •H I J, N+J l -TAU/PI J) *VI NS+J !·SUM! Y, NS ,A, UG.I', VI N6+ 1) ,J, Nl
HINP I,J l•ISI•AI NPZ ,I l•SZ*AI NPZ ,NPZ l l"YI N6+ I l·ALL"SZ
HINPl,N+Jl•IAINP2,JP1l+S3*AINPZ,Il+S4*AINP2,NPZll 0 YIN6+1l•ALL*S4
AAIJ,NP1l•YIN+Jl•AIJPl,NPZl
11• CONTINUE
AA<NPl,NPl>•V(HG+S)•(AtNPZ,N,2>•TAU/P(NP1))-SUH(V,H.A,V("S+4),
YIN6•5l ,NPI ,Nl
CALL 8NDINV IAA,NPI,ITESTI
CALL HATCPY IAA,F,75,75,Z.I',2.,NPI,NPll
CALL HATMULT(F,H,2•• 2•• Z.,4.,NP1,N2,NP1l

Table A4-7 Continued

 -446-

C CAlCUlATE THE I MATRIX

ST•IYIN6+4l-I.D.8'l/AL;/VIM6+4l
DO 11.8' 1•1, N
I P 1., I +1
S T l .. VI N$ • I ) • 'A ( I P 1 • \ >•$ ( 3 , 3) •A ( I P l • MP 2) •s ( .. , 1) )
RI I I , 1 l • • YI N5 •I l *SUM I Y. N,A, VI N6+4 l, Yl N6+5 l, l , N)
+VIN+Il*SUMIV,N5,A,UG.8',VIN6+ll,I,Nl+TAU/P!Il*
YINS+ll*YIN+ll+STI*ST
Rll 1,2>•-AI !Pl.! l*VIN+I l-STI*ST
RIII,Jl•-STI•ST
Rlll,4l•STI/ALG/YIN6+4l
116 CONTINUE
STI•YIN6+1l*IAINPZ.ll 0 SI3,3l+AINPZ,NPZl"SI4,3ll-ALl*SI4,3)
Rl<NPl,l>•-Y<N6+1J*SUMtY.N.A,Y<N6+,),Y{H6+5),NPl,N)+V(N6+5)•
SUHIY,N5,A,UG.8',VIN6+ll,NP1,Nl+TAU/PINP1l*YIN6+ll 0
VI N6+5 l •ST1°ST
RIINPI.2l•-AINP2.1l*VIN6+5l-STI*ST
R11NPI.3l•-ST1•ST
Ri\~P1,41•STlfALG/V(N0+.)
CAI.L MATHULTIF,Rl,Z.IJ,Z6,2.8',4,NP1,4,NP1l
C FINAI.LV SET UP THE STATE MATRIX 'A'.
CALl. ZEROMIAA, 75,751
CALL ZEROMIBB,7S,Bl
CALL ZEROVIDD,l~J
AB•I2.D.8'•VIN6+4ll/ALG/YIN6+4l
DO 25.8' l•l,N
I l• I+ 1
STT••AB•ISI3,3l*Aill,ll+SI4,3l*Aill,NPZll
ST l•A I 11 , 1 l -A I NP2, I l •AI 11, NP 2 l I A I NPZ, NP2 l
ST2•STl•SC02B
ST3•.8'. DB
DO 199 K•I,N
STT•STT+QIZ,I.Kl*VIKl+QI4,l,kl*VIN+Kl
ST1•STI•IAII1,K•ll-A1li.NPZl"AINP2,K+Il/AINP2,NPZll"YIN3+kl
ST2•STZ+IA1li,K+ll-AII1,NP2l•AINPZ,K+ll/AINPZ,NP2ll*VIN4+Kl
t•• ST3•ST3+Qil.l.Kl*V(Kl+QIJ,I,Kl"V(H+Kl
DO Ul J•I,N
Jl•J+l
AA\ I,J>•A~S*I8( li,JI l+BI ll,ll*ISI 1,1 l 0 AI I ,JI )+$1 1,2l*AINP2,JI ll
+81li,NP2l*l$(2,1l*AII,Jll+SI2,Zl*AINPZ,Jilll
AAC I ,J•Nl•A~S*I8( II, I l*l$1 1,3l*AI I,Jl >•SI I, A l*AINP2 ,JIll
•8CI1~NPZJ*<St2.3l*A(l,Jl>•S<Z~4>*A(NP2,J1JJ)
AA I N+l, J l •Q I 1 , I , J l •v I N+l l +QI 2, I, J l •v I HS•I l •H I I , J l •STT
AAI N+! ,N+J l•QI J ,I ,J l*VIN•l )+QI 4, I,J l*VI NS+I l+HI l ,N•J l*STT
AAIN2+1,N2+Jl•ALT*IBII1,Jil-BII1,NPZl*AINPZ,Jil/AINP2,NPZll
AAIN3•I.Jl•-HII,Jl*STI
AACN3+l,N+J>•-H(I,J+N)•ST1
AAI N3•1, NJ+J l •-VI NS+I )• IAI II,Jl I-AI NPZ ,J I l*AI 11 ,NPZ l /AI NPZ, NPZl)
AAlN4+l,JJ••Hti.J)*STZ
AAIN4+I,N+Jl•-HII,N+Jl*ST2
AAIN4+I,N4+Jl•AAIN3+1,NJ+J)
2.8'1 CONTI HUE
S1•1.0.8'+PHIZ*Yill
S2•l.DB+PHI3*VIll
S3 •B Ill • I l •s I 1 • 3 J +8 I II,"' Z l •$ I Z, J l
S4•8111,1l*SI3,31•BII1,NP21*SI4,Jl
SS•AIII,ll*S(3,3l+A(I1.NP2l*SI4,31
S6•VIN+ll/PII l
AAI I, I l•AAI I,l !+Ill 1 l+OEI*Ikll I I,2l*SI+PHI2*RLlll, 1 l l+DEZ*
IRL211,2l*SZ+PH13*RL21l,lll
AAI l,N+I l•AAII,N•I l+QAS
AAI I,NZ+I l•lll2l
AA1l,N3+ll•DE1*RLIII,3>*SI+DEZ*RI.211,3l*S2
AAI1,N4•Il•DEI"RL11!,4l*Sl+DEZ•RLZI!,4l*SZ
AAIN+I,Il•AAIN+I,Il+GAG•$6
AAIN+I,N+Il•AAIN+I,N+ll+ST3+(($(3,31*81li,Il+SI4,3l*SIII,NP2ll
*AB*ALG+Il16l*PH3•GAG*YIIl+ll1Sl*VIN2•Il•2.0••Ill4l*YIN+lll/PIIl
AAIN+I.HZ+Il•ll15l*S6
AAINZ+I,I !•Ill 71
AAIN2+l,N+Il•WI8l
AAI N2• I, NZ+I) •AAI NZ•I, NZ+l l +Ill 9 l
AA I N3+1 , I l •AA I Nl+ I • I l +IS I Z*RL 2 I I, 2 l •S II*U 1l I, 2 l l *S6
AAI NJ •l , N<I l •AA I N3 +I , H+l l +IS I2*RLZ I I , 1 l -S II •RL 1 I I, I l l /PI I)
AAINJ•I,N3+ll•AAIN3•l.N3+1l+ISI2*RI.ZII,3l-SII*RL111,3li*S6
Ul N3+l ,NHII•!SI2*Rl2! l. 4l-Sli*RI.1 { I.4 l l*$6
AAI N4+I ,I l •AAI N4+I, I l -S IJ"IUZI I ,2 l*$6
AAI Nhi,N•I l•M! U+I,N+I l-SI3*RLZI 1,1 l , l I l
AAIN~•I,N3+ll•-SI3*~LZII,3l*S6
AAIN4+I,N4+ll•AAIN4+l,N4+Il-SI3*RL21!,4)•$6

Table A4-7 Continued

 -447-

IBCI,Il••ST*ALS"$3
IB! I ,Zl•U! I, I l
88!!,3l•ALS*S3/ALG/YCN6+4J
DO 31611 J•4,8
BBCI.Jl•DEI*RLICI,J+ll*SI+OEZ*RLZ!I,J+ll*SZ
98{ N3+ I ,V }•( SI:Z:.,.RlZt I ,\1+ 1 > ··Sll ~1\L 1 (I ,V+l) ).,..$6
3.11 BBCN4+l,Jl•·SI3*RLZ<I.J•ll*S6
BBCN+I,Il•-VCN+Il*ST3·CV!4l*Y!N+Il+VC5l*Y!NI+Il•GAQ*V!II+VC&l*
PH3•<AB•STI*ALG*S4l*S6+ST*YCN5+ll"S5•RI!I,3l*STT
BB!N•l.Zl•R!Cl.21*STT·ST*S6*AL~*S4•ST•YCN5•li*S5
B8CN+l,31•CS4*S6·Y<NS+II*SS/ALGI/Y!N6+41+RI!l,4l*STT
BB<N2•!.3J•ALT*CBC!1,11•8(11,NPZI*ACNP2,1J/ACNPZ,NPZII
BB~NJ+l.lJ•-Rltl.:JJ•~tl
88CN3+1,Zl•·R!CI,21*STI
BB<N3+1,3J•·RICI,4l*STI
BB!N3+1,4l•B8CN3+1,4J-Y!NS+II*(ACI!,II-A!NP2,11*A!Il,NPZJ/
A<NPZ.NPill
881 N4+1,1 l•·R I< I, 3l*ST2
88! N4•1, 2) •·R I (I, Z l *ST2
88!N4+1.31•·~1<!,4l*STZ
88!N4+1,5J•BBCN4+1,SI·YCN5+1l*CACII,ll•ACNPZ,Il*A!II,NP2l/
AINP2,NP2l I

ALIII•!•CPGI*M;*ST.+!.77570•3J*ST.+&.775l/MG/CPG
AL(2l•CP~I•ST./CP~
AL13l•SCO*IST11*!.4Z430-3l-.1750111/MG/CPG
Al!41•SCO*ISTI6*!3.441630-31•l.Z76Z60161/MG/CPG
Al!51•SCO*ISTII*!6.39430-31+4,4250161/MG/CPG
AL!61•SCO*!STI6*!,163430-31·.155016J/MG/CPG
AL!7l•SCO*<STII*!11.65Z50·31·1.57420111/MG/CPG
$T4•Rll{I.z>•Vtil+RLlfl.3>•V(H3+ll•RLl(l,,>•Y(M,•l>•RL1(1~1>
+RLl!1,61*SCOZB+RLl!l,71*SH20B+RL1CI,81*SHZB•RLlll,,:
*SCH.S
ST5•RLZ!l,ZI*VIII•RLZI!,31*Y!N3+ll+RL211,4l*Y!N4•li+RLZ1l,51
+RL2!l,61*SCOZB•RL211,7l*SHZOB•RL211,81•SH28
DDt I l•DEI*Sl*IRLI I I ,I l·SH I·OE I*RLI I 1,1 l*PHI2*V! I I
+OE2*SI*!RL21l,ll-STSI·OEZ*RL21l,ll*PHI3*YIIJ+ST•ALS*S3
+ALllJ•t8<I,l>
DOIN+Il•STT*!RIIl,ll·VINS+III+ST•tAL;•s&*S4·SS*Y!NS+Ill
+AL!ll*BBIN+I,IJ
DDINZ+IJ•S.DS
DDINJ+IJ•<VINS•II-RIII,Ill*STI+Sil*S6*ST4·SI2"S6•ST5+
•AL!II*BB!N3+1,11
OD<H4+li•IV(NS•II·RIII,lli*STZ+SI3*S6*STS+AL!ll*BBIN4+1,1)
2516 CONTINUE

DO 3516 I•I,HS
DO 3516 J•3,8
35• BB<l,Jl•BB!I,Jl+ALIJ·ll*BBII,II
()O •1111 I•I.N
BB!1,ll•V13l
8B!N+!,II•V16l*YIN+Il/PIII
BBINZ+l,ll•VIl.l
IB!N3+1, 1 1••• 016
••• sa<N4•I,tl••.o•
ltETUitN
END

Table A4-7 Continued

 -448-

SUI~O~!ME ACTl$$(U,V.E,,.l
C THIS SUBROUTINE CALCULATES THE FINAL STEADY STAT£
IMPLICIT RElL"B !A-H.O·Zl
REAL 0 8 HG.KCZB,KPZA.KP2B.KlP,KZP
DIMENSION U18l,Y125J,VIlBBJ
COMMON /GASPA/ CPG,PTZ,PTR,~,PQS,UeS,UM
/OPCON/ SCH4,SCO,SCOZ,SH2,SH20,PTI,PT!,STB,STV
/REACZ/ EXPA.EXPB.EXPC,KCZB,KPZA,KPZB,KlP,KZP,EAI,EAt
EXTERNAL FNl
ST•HG
VI IBJ•STV
ST'W'•U( 1 >•ST6
UGS•U!Zl"UGS
ST.IJ•V13l 0 STB
SCO•UI ' l 0 SCO
SC02•Ut5l"SCO
SH20•Ut6J 0 SCO
SH2•U!?J•sco
scH••uteJ•sco
CALL CPCALCtSH2,SCO,SC02,SH20,SCH4,STB,H;,cpG,CP;I,CP;zJ
PGS•PGS'MGISTIUC3J
KJP•KlP 0 UI4l
1<2P•K2P•U{41
DO !U 1•1.8
VI l J•UI l l
IBB UIIJ•I.DB
UI I J •STII /STJ
UISJ•SCOZ/SCO
Ut6J•SH2C/SCO
Ut7J•SHZ/SCO
UI 8 l •SCHVSCO
Vl9l•UM
UH•l.OB
CALL DIHLES
CALL RADIAL
CALL NSES116°N+5,Y,EP,1 •• ,g,FNll
DO lBI 1•1,8
1Bl VI I l•Vt I J
UH•VI9J
DO 112 1•1. N
VI I J•YI I l"Ut3J
Y(N•I )•VtM•I>*U(3l
V(2*N•J )•VC:t•N•!)•Ufl)
V<3*N+J)•V<l•N+t)•UC4)
V(4•N•l)•V(4•N•ll*U(4)
1•2 YI5"N•ll•VI5"N+ll 0 UI2l
VO.i'*N+1l•V<6*N+l >•UtZ>
YI6•N•Z)•Vt6•N•t>•Ut3>
Vt6•N+J)•V<6*M+3)•Ut3>
V(G•N•~)•V(6*N+4>*U(3)
VI6*N•SJ•Vt6*H•5l"Ut3J
ST•Ht;
ST\l•V( lBJ
UCS•UGS/UtZl
STJ•STB/U( 3l
SCO•SCO/UI41
$COZ•$COIU<5l
SH20•SCO/Ut 5J
SHZ•SCO/Ut7l
SCH4•SCOIU18l
CALL CPC4LCt$H2,SCO,SCOt,SMZO,SCHI,ST•• MC.C,C,C'Gt,C,G21
P'S•P;S•"(i!ST 0 Ut3l
KlP•KlPIUC•l
KZP•KZP/Ul"J
CALL DIHLES
CALL -ADIAL
RETURN
UD

Table A4-7 Continued

 -449-

SU&~OUTINE INSIMIY,YS,U,N,~FLA;)

C THIS SUBROUTINE INITIALIZES THE SIMULATION.

IMPLICIT REAL*& IA•H,O-Zl
REAL•& HC,MGB
DIMENSION UtB>,VtJBBl,YS(IBBl
COMMON /CASPA/ CPC,PTZ,PTR,MC,PGS.UGS,UM
/OPCON/ SCH4,SCO,SCOt,SH2.SHZO,PT.,PT1.3T.,STW
/LALCB/ St~,4l,Ht2B,4Bl,Rit2B,4>,ALI7l

llEAOtS,!l JFLAG
IFIJFLAG .EO. I >CALL INITIALIY,U,Bl
JFtJFLAC .EO. llRETURN
DO IBB I•I,S"N
lBB Ytll•YSIII
READ 15,2> UCB,TB,XCO,XC02,XH20,XH2,XCH4,TV
UtI >•TV/STB
UIZI•UGB/UCS
Ut3l•TB/STB
Ut ' l •XCO/SCO
Ut5>•XCOZ/SCO
UI&>•XHZO/SCO
Ut7l•XHZISCO
Ut8l•XCHHSCO
UM•ALtll+ALIZl*U13l+Alt3)•Ut4l+ALI4)*U(5l+ALISl*Ut6l+AL16)
*UC7)+Alf7l•U!9l
PRINT 3, !UtI> ,1•1,8> ,UM
~ETURN
1 FORMATtl2>
Z FORMATI9F8.21
3 FORMATI/1/,ZX,'INITIAL CONTROL VECTORo',//SX,'TV •',F8.4,1X,
'UGB •',F8.4,SX,'TB •',F8.4,5X,'XCO •',F8.4,5X,/5X,
•)((';02•'. Fl. 4, SX. 'XW20•' ,Fa. 4, SX, 'XH2 • •. ~8. 4, SX. 'XCM4••.
F8.~/5X,'UM •',F8.4>
END

SUBROUTINE OUTCALC!U,Y,VS,ALG)

C THIS SUBROUTINE CALCULATES THE VELOCITIES AND ENOPT TEMPEilATURES

IMPLICIT REAL*& IA·H,Q-Z>
REAL•a M;
DIMENSION YI1BB>,SAI4l,Ut8>
COMMON /LALWB/ S< 4. 4) .H( Z6.4•> .Rl ( Z6,4) ,ALt'tl
/COMAT/ AIZS.ZSI ,BtZ5,Z5> ,RtZ5l ,N
t;ASPA/ CP;,PTZ,PTR,MG,P;s,u;S,UM
NP2•N+2
MS•N*S
N6l•N*6+1
CALL ENDPTS (TTB,Yl.,YlNPl,YZ.,YZMPl,TTNPl,Y.UV.)
DO IBB I•I,N+I
VI NS+I ) •R Ill ; 1 I +RIll , Zl*U~B+R I I I , 3 >*UH+R I! I, 4> *U t 3 l
DO ll!l J•l.N
IB1 YCNS+ll•Y!NS+I)+Htl,Jl*YtJl+Hil,J+Nl*Y(N+J)
IB6 CONTINUE
CALL ZEROVt$A,4l
DO In J•I,N
JI•J+I
SA< I >•SAtll+A!I,J1l*YI~)
SAtZl•SAIZl+AtNPZ,Jil*Y<Jl
SAt3)•$A13l•AII,Jil*YtN+Jl
1•z SA14)•SA<~l•ACNP~.~1)*V(H+J}
ST•IYS·I.OBl/ALG/YS
SAt3l•ST-ST•UM-ST•u;B•UI3l/AL;/YS+SAI3l
DO IB3 1•1,4
Y!N61+1 >•B.Ol!
DO U3 J•l. ~
l.J Y< nti 1•1 >•Yt N61•l )•S( 1 ,V >*SAC.,)

RETURN
END

Table A4-7 Continued

 -450-

SUI~OUTINE CPCALC!SH2,SCO,SC02,SH20,SCH4,ST.,MG.CPG,CPGI,CPG2l
C THIS SUBROUTINE CALC$ THE HEAT CAPACITY
IMPLICIT R£AL*8<A~H.O•Zl
REAL*S M~
SNZ•l.DB·SH2·SCO·SC02·SH20-SCH4
MG•SCH4*16.B43DB+SC0*28.1114DB+SC02*44.1.98DI+SH2*2 •• 16DI
+SH20*!8.B154DB•SN2*28.1!3401
CP~I•!SH2*.810·3•SN2*.77570·3+SC0*1.20-3+SC02*4.Zl60·3
~SH20•?.l?0-3+SCH•*12.42B20-3}/HC
cP;2•<SH2•6.&2og•sH2*6.77So••sco•s.&o••scoz•e •.stzso•
+SH20*11.20B+SCH4*S.ZII801l/MG
CPG•CP;t•STI+CPGZ
RETURN
END

SUBROUTINE ENDPTS <TTI,Yl.,Y1NPl,YZI,Y2NP!,TTNP1,Y,UGil

C THIS ROUTINE CALCULATES THE ENDPOINT CONDITIONS.
IMPLICIT REAL*8 <A·H,O-Zl
DIMENSION V\ l8'JI'l
COMMON /COMAT/ A!25,25l,8(25,25l,R!25l,N
ll !NEA/ AA! 75.75 l .UI75, 8l. 00! 75 l ,U! 8l
/D!MLE/ ALS,ALG,ALT,BES,BEG,GAS,;AG,GTS,GTG,OE!,
D£2,SI!,SI2,Sl3,PH!2,PHI3,PHI,PHZ,PH3
NP2•N+2
NZ•Z•N
N3•3*N
N.(•.t•N
UGB•UCZ>•Y{N•6•4J/Ut3>
V11•U!4l
V28'•U!Sl
VlNP1•VtN•6+6)
Y2NPl•·A!NP2,1l/A!NP2,NPZl*U!5l
TTNPI•B.OI
TTI•U!Jl
00 liB I•I,N
I P 1•1 • l
V2NPl•V2NPI•AtNP2.1Pll/AtNP2.NPZl*Y!N4+1l
TTHP!•TTNP!+A!HP2.1Pil*Y!N2+ll
1•1 CONTI HUE
TTNPl•-ITTNPI+AINPZ,Il*TTBl/A{NP2,NP2l
RETURN
END

FUNCTION SUM!Y,NN.AA,XI,XZ,I,Nl
IMPLICIT REAL*8 !A-H,O-Zl
DIMENSION Yl liB) ,AA<ZS,ZSl
SUM•g,.og
l p 1•1 +!
00 !U J•I.N
t•• JPI•J•I
SUM•SUM+AAtiPI,JPil*Y(J+NNl
SUM•SUM+AAI IP 1,! )*X I+AA( IP l,N+Z l*XZ
~ETURN
£NO

SUBROUTINE REAC (VI, YZ, TS, TG,PT ,RIP ,ltZP l

C THIS SUBROUTINE CALCULATES THE DIMENSIONLESS RATES
IMP.ICIT REAL*8 (A·H,O-Zl
REAL*8 MG,KPZA.KP28.KC21,K!P,K2P,KP2
COMMON /OPCON/ SCH4,SCO,SC02,SH2,SH20,PTI,PTI,STB,STV
/REAC2/ EXPA.EXP8,EXPC,KC21,KP2A.KP28,KIP,K2P,EAI,EA2
/LINEA/ AA<75,7Sl,8Bt75,8l,00<7Sl,UI8l

GA•B.D8'
TT•STI*TS
KP2•DEXP<KP2A•KP28/TTl
TH!•SCO*!U!4l·Y1+U(5l•Y2l
TH2•SCO*IU!Sl•Y2l
TH3•l.DI·2.D••TH!
VH2•1SCO*U!7l·3.0B*THI·TH2l/SCO/TH3
VHZO•!SCO*UI6l+TH!+TH2l/SCO/TH3
YCH4•1SCO•U!8l+TH!l/$CO/TH3
VCO•YI/TH3
VC02•V2/TH3
PA•PT*TH3
!f!yh2 . H. ••• dllyh2•1.d·11
!ftyco .lt. B.dllyco•l.d-11
k1P•OEXP(-£A1/TT>•PA*•texp••expb)*VHZ••expb*YCO**expa/
(l.OB+KtF•PA•VCO+KZP•PA*YHZ)*•e~~c
R2P•DEXP<·EA2/TTl*!.8301+.!7D••PA*PT•>•<VC02*YH2-VCO*VHZO/
KP2*KC21l
llETURN
END

Table A4-7 Continued

 -451-

FU~CT!O~ r.cT,TT,L,Rl
IMP~lCIT R£AL•9 (•-w.0-71
COMMON /RADIAl V!llJl,VPt9,Zl,D£TAIZJ,RR
/LINEA/ AA<?5,7Sl,I8175,81,0017Sl,Uiel
TR• I !VP { t,l )+VP!., Ll*R•VP I 7 ,L l 0 R**2 l*T
• !VP 12 .l 1+1/P I 5, L l*R+IIP I 8,l J•R••z l*TT
+VP <3,LJ•U! I 1•VP I 6,li*U! II•R•VP! 9,L J*UI I l*R .. %1/0ETAI L 1
UTU~N
END

FUNCTION RFN IAA,NNl

IMPLICIT REAL*8 !A-H,O-Zl
OIMENS ION AAI llJlJ)
ST•lJ. 0.111
00 I.U I•!,NN
llJlJ IF' DA8S IAA I I )) . fOE. STI ST•AA!I l
RFN•-DLO;!lJ!DABSISTll
RETURN
END

SUBROUTINE COLLOC
C THIS ~OUT!NE CALCULATES THE ZEROS OF THE ORTHO;ONAL POLYNOMIAl AND
C S£TS UP THE AXIAL COLLOCATION MAT~!C£S A ANO 8
IMPLICIT REAL*8 IA-H,O-Zl
DIMENSION OI!Z5J,OZ!Z51.03!Z5l,V1!25l,VZ!251
COMMON tCOMAT/ A!ZS,251,8!25.251,R<Z5l,N
CALL JCOBIIZ5,N,I,t,lJ.DlJ,lJ.DlJ,Ol,OZ,D3,Rl
PRI~T 11 <k{J>.J•l,N~Z>
00 U! l•I,N+Z
CALL OFOPR<.S,N,l,l,l,l.DI,DZ.DJ.R,VIl
CALl DFOPRIZS, N,l,!,l,Z ,OI,OZ.D3,R, VZ l
DO U2 J•!,N+Z
A!I.JJ•VltJJ
BII,JJ•VZIJJ
llJZ CONTINUE
llJI CONTINUE
UTURN
FORMATI/////,3X,'AXIAL COLLOCATION ,OINTSt',//,3!1~,41EIZ.41/))
END

SUBROUTINE tHTLSSIV,U,IFLA,,HFLA~,L,U;S,PHI,N.EPl

C TMIS kOUTINE CALCULATES TME INITIAL STEADY STATE P~OFILE

lHPLICIT REAL•S !A~H,O-Zl

r••1•8 1
DIMEHSION VlllJlJJ,U(Sl
EXTERNAL FNI,FHZ
CALL INITIAL IV,U,!FLA~J
CALL OUTPCTIV,U,lJ.OlJ,L,U~S.PHl,IJ
IFINFLA' .EO. IJCALL NSES1!6•N•6.V,EP,ZlJlJlJ,lJ,FNI)
IYINFLAG .!0. ZlCALL N3ESZ!6°N+5,V,EP,5BlJ,!T,FN2J
!FliT .EO. SlJ.IJJ PRINT I
CALL OUTPUT!V,U, •. DlJ,l,U;S,PHI,Zl
RETURN
FORMATI//I,ZX,'REACHED MAXIT IN WSESZ'//1
I liD

Table A4-7 Continued

 -452-

APPENDIX5

ORTIIOGONAL COlLOCATION ON F'INITE ElEMENTS 110DEL

The technique and application of orthogonal collocation on finite elements

is described in detail in Section 3.5. This appendix presents a summary of the

final coupled system of algebraic and differential equations. Computer pro-

grams for solution of the packed bed reactor model using orthogonal collocation

on finite elements are presented in Tables AB-1 and A5-2 and are stored in direc-

tory [RRK.OCFE]. These are very similar in structure and operation to those in

Appendix 4.

Arter applying orthogonal collocation on finite elements, the packed bed

reactor model becomes

Catalyst Energy Balance

For elements 1c :;: 1. ... , NE,

within each element 1c for the interior collocation points I :;: 2, ... , NA:+ 1. For ele-

ments k :;: 2, ... , NE,

and for the first (k :;:1) and last (k :;:NE) elements:

 -453-

Gas Energy Balance

For elements k =1, ... , NE,

within each element k for the interior collocation points I= 2, ... , N,~:+l. For ele-

ments k = 2, ... , NE,

and for the first (k =1) and last (k =NE) elements:

 -454-

.Mass Balances

For elements k = 1, ... , NE,

+ 4yl/
(1-2o) 2
[ !! l2] +
v ..

where

within each element k for the interior collocation points I= 2, ... , NA:+1. For ele-

ments k = 2, ... , NE,

and for the first (k = 1) and last (k =NE) elements:

 -455-

Thermal Well Energy Balance

For elements k = 1, ... , NE,

within each element k for the interior collocation points

I = 2, ... , N.t + 1. F'or elements k = 2, ... , NE,

and for the first (k =1) and last (k =NE) elements:

atl = 6to
II
~ Av.NE·" et1
./=1
= 0

Overall Continuity

For J = 2, ... , N-"+1 for each element k = 1. ... , NE-1 and for J = 2, ... , NNE+2
fork = NE:

For elements k =2, ... , NE,

1
= [ -h ~ Auv&J]
.1=1
lc: ~lc:

and fork =1
 -456 ~

C THIS PROGRAM MODELS A NONISOTHERMAL. NONADIABATIC FIXED BED REACTOR

C WITH BOTH A COOLING JACKET AND A THERMAL WELL. THE ANALYSIS IS
C PERFORMED FOR A HETHANATION OF CO SYSTEM.
C THIS PROGRAM USES THE ORTHOGONAL COLLOCATION OF FINITE ELEMENTS METHOD
C LINK TO MLIB,OCFELIB
C THIS PROGRAMS SOLVES THE FULL NONLINEAR HODEL WITH THE CONCENTRATIONS
C NONDIMENSIONAL W.R.T. THE STEADY-STATE INLET CONC. OF CO
IMPLICIT REAL*B <A-H,O-Z>
REAL*B L,KT,HG,KS,KSP,KPlA,KPlB,KC1S,KPZA,KPZB,KCZS,KZP,K3P
COMMON /REACP/ EPS,L,RS,Rl
• /CATLS/ CPS,PS,TC,DC
* ITHWEL/ CPT,PT,KT
* /GASP A/ CPG,PTZ,PTR,HG,PGS,UGS,UM
* /HEATT/ OHSG,OHTS,OHTG,BGS,BSG,BTS,BTG,BWG,BWS
* /OPCON/ SCH4,SCO,SC02,SH2,SH20,PTS,PTl,STS,ST~
* /REACl/ DH1A,DH1B,DH2A,DH2B,KS,KSP
* /DIMLE/ ALS,ALG,ALT,BES,BEG,GAS,GAG,GTS,GTG,DEl,
* DE2,Sil,SI2,Sl3,PHI2,PHI3,PHl,PH2,PH3
/RADIAl WI1S>,WP19,2>,DETA!2>,RR
** /COMAT/ NE,N<S>,H<5>,ZII6>,NSI6>,Z<S,ZS>,AI5,25,25),
* 8!5,25,25>
* llNDEX/ NSE,NTS,NTT,NTG,NCO,NCZ,NUG,NNE,HEl
* /HISCl/ UU< US>
* /REACZ/ KC1S,KP1A,KP1B,KC2B,KPZA,KPZB,KZP,K3P,EA1,EAZ
* /LINEA/ AA!75,7S>,BBI7S,B>,DD!7S>,U<8>
* /STATE/ V!lSS>,YDOT!lBS>,ST,DT

EXTERNAL DERIV,FN

C READ IN DATA AND CALCULATE CONSTANTS

CALL SETUPS <DTS,TMAX,DL,RR,EP,NP,IF,IFLAG>
C CALCULATE THE DIMENSIONLESS PARAMETERS
CALL DIMLES
C CA~CULATE COHST~HTS FOR THE ~AD!AL LUMPED MO~EL

CALL RADIAL
C CALCULATE ZF.ROS OF THE ORTHOGON~L POLYNOMIAL AND SET UP AXIAL
C COLLOCATIO~ MATRICIES A AND B.

CALL C~!.lOC

C SOLVE FOR THE STEADY STATE PROriLE

CALL INTLSS!V,U.FP.IF~A~l

Table A5-1
Computer Program cx.:.f'E
 M 457 M

SUBROUTINE DIMLES
C THIS SUBROUTINE CALCULATES THE DIMENSIONLESS ,ARAMETERS
IMPLICIT REAL*8 IA-H,O-Zl
REAL*B L,KT,MG,KS,KBP,KCZO,KP2A,KP28,K1,,K2P
COMMON /REACP/ EPS,L,RB,RI
• /CATLS/ CPS,PS,TC,OC
/THWEL/ CPT,PT,KT
*
* /GASPA/ CPG,PTZ,PTR,MG,PGS,UGS,UM
/HEATT/ OHSC,OHTS,OHTC,8CS,8SC,8TS,BTC,8WC,IVS
•* /OPCON/ SCH4,SCO,SC02,SHZ,SH20,PTB,PTI,STB,STW
/REACt/ DHIA,DHIB,OHZA,OH2B,KB,KBP
*
• /OIMLE/ ALS.ALG,ALT,BES,BEG,GAS,GAG,GTS,GTG,DEI,
• DEZ,Sll,SI2,SJJ,PHIZ,PHI3,PHI,PH2,PH3
* /REACZ/ EXPA,EXPB,EXPC,KC2B,KP2A,KPZB,KIP,KZP,EA1,EA2
DATA PI /3.1415!Z654DB/
VT•Pl*RB**Z*L
VB•PI*L*IR1**2-RB**2l
AZ•LIDC
AR•Rl/DC
AO•L/Rl
EMl•l . .B'OB-EPS
ALS•TC/IPS*CPS*EMl*L*UGS>
ALG•l.DB/IPTZ*EPS*AZl
ALT•KT/IPT*CPT*L*UGSl
BES•ALS*L**Z/Rl**Z
BEG•AO/PTR/EPS/AR
GAS•OHSG*L/IVB*PS*CPS*EMl*UGSl
GAG•OHSG*L/IVB*PGS*CPG*EPS*UGSl
GTS•OHTS*L/IVT*PT*CPT*UGSl
GTG•OHTG*L/IVT*PT*CPT*UGSl
DEI•-L*OHIB*IPTB*SCOl**IEXPA•EXPBl*KBP/IUGS*CPS*STBl
DE2•-L*OH2B*SC0**2*KB/IUGS*CPS*STBl
SII•MG*PS*EMI*IPT.B'*SCOl**IEXPA•EXPBl*KBP*L/IEPS*UGS*PGS*SCOl
SIZ•MG*PS*EMI*SCO*KB*L/IEPS*UGS*PGSl
SI3•SI2
PHt2•0HIA*ST.8'/DHIB
PHI3•0H2A*STB/DH2B
8\JG•BTG
B\.IS•BTS
PH I•R.B'/R I
PH2•PT1/PTIJ
PHJ•STW/STB
RETURN
END

Table A5-2
Computer Program ocm..IB
 -458-

SUBROUTINE INITIAL IY,Ul

C THIS ROUTINE INITIALIZES THE SYSTEM
IMPLICIT REAL*B IA-H,O-ZI
II.EAL*B MG
DIMENSION Ylllllll,UIB)
COMMON /OPCON/ SCH4,SCO,SCOZ,SHZ,SH20,PTJI,PT1,ST6,STW
* /GASPA/ CPG,PTZ,PTR,MG,PGS,UGS,UM
* /COMAT/ NE,NISI,H15l,Zll6l,NSI6l,ZIS,ZSI,AI5,25,25l,
• 815,25,25)
/INDEX/ NSE,NTS,NTT,NTG,NCO,NCZ,NUG,NNE,NEI
DO 366 1•2,4
3.. Ulli•I.D.
UI I l •ST\USU
U!Sl•SCOZISCO
Ul6l•SH20/SCO
Ul 71•SH2/SCO
Ul8l•SCH4/SCO
UM•1. 011
II.EADI5,11 TC,TS,TT,Xl,XZ,VEL
DO IJJII K•l, NE
CALL INDICESIK,NK,NI,NZ,N3,N4,N5,N6l
DO !JIB l•l,NK
YINI+I>•TS/STJI
VIHZ+II•TT/STJJ
YIN3+ll•TG/STJI
VIN.t+II•XI/SCO
YINS+Il•X21SCO
YIN6+ll•VEL/PRESil,Kl/UGS
1•11 CONTINUE
DO ZBII J•I,HEl
VINTS+JI•TS/STJJ
YINTT+JI•TT/STJJ
VINTG+Jl•TG/STB
YIHCO+JI•Xl/SCO
Y<NC2+Jl•X2/SCO
IFIJ .HE. NEllVINUG+JI•VEL/PRESIII,JI/UGS
IFIJ .EQ. NEIIYINUG+JI•VEL•PT6/PT1/UGS
2611 CONTINUE ,
RETURN
F'ORMATI9F'8,U
END

Table A5-2 Continued

 -459-

SUBROUTINE SETU,S IDT.,TMAX,DL,RR,E,,N,,IF>

C THIS SUBROUTINE READS IN THE INPUT DATA AND MAKES PRELIMINARY
C CALCULATIONS.
IMPLICIT REAL*B IA-H,O-Zl
REAL*B L,KT,MG,KB,KBP,KP2A,KP2B,KC2B,KIP,K2P,K2,K3
COMMON /REACP/ EPS,L,RB,RI
* /CATLS/ CPS,PS,TC,DC
/THVEL/ CPT,PT,KT
* /GASPA/ CPG,PTZ,PTR,MG,PGS,UGS,UM
* /HEATT/ OHSG,OHTS,OHTG,BGS,8SG,BTS,8TG,8WG,BWS
* /OPCON/ SCH4,SCO,SC02,SH2,SH20,PTB,PT1,STB,STW
/REACI/ DHIA,DHIB,OH2A,DH2B,KB,KBP
* /REACZ/ EXPA,EXPB,EXPC,KC2B,KP2A,KP2B,K!P,K2P,EA1,EA2
* /COMAT/ NE,N15l,HI5l,ZI16l,NS<61,ZI5,Z5l,A15,Z5,251,
* 815,25,251
DATA RG,RGP,PI /82.BS~~DB,t.987DB,3.1~15!12654DB/
kEADIS,Il EPS,L,RB,kl
AEAOIS,ll CPS.PS.TC,DC
READIS,Il CPT,PT,KT
READIS,Il PTl,PTR,UGS
READIS,ll OHSG,OHTS,OHTG,BGS,BSG,BTS,BTG
READ<S,ll SCH4,SCO,SC02,SHZ,SH20,PTB,PT1,ST.,STW
READIS,ll OHIA,DHIB,DH2A,OH2B,KB,KBP
READIS,ll EXPA,EXPB,EXPC,KP2A,KP28,K2,K3,EAI,EA2
READI5,2l DTB,TMAX,RR,DL,TS,NE,NP,NEP,IFLAG,IF
REA015,6l INIJl,J•l,Sl,CHIJl,J•l,Sl
CALL CPCALCISH2,SCO,SC02,SH20,SCH,,ST.,MG,CPG,CPGI,C,G21
PGS•MG*PTB/RG/ST.IJ
K1P•K2*PTB*SCO
K2P•K3*PT."SCO
KCZ.IJ•l.DB
EP•IB.OB,..NEP
KBP•KBP*TS**I-.31
PRINT 3,EPS,CPS,CPT,L,PS,PT,RB,TC,KT,Rl,OHSG,DHIA,EXPA,
* OHTS,OHIB,EXPB,OHTG,DH2A,EXPC,BGS,DH2B,KP2A,BSG,Kr,
* KP2B,8TS,KBP,K2,8TG,K3,PTl,EAt,PTk,EA2
PRINT 4,DTB,TMAX,DL,RR,TS,EP,OTB*4.0B**IIF-11
PRINT 5,SCH4,STB.SCO.STW.SC02.PTB.SH2,PTI,SH20,MG,UGS,CPG
OT.IJ•DTB*UGS/L
TMAX•TMAX*UGS/L
DL•DL*UGS/L
RR•Il.DB-RB/Rll*RR+R.IJ/RI
EAI•EAI/RGP
EA2•EA2/RGP
RETURN
l FORMATI!ID8.21
2 FORMATI5F8.2,/6IBl
3 FORMATIIH1,////,3X,'INPUT DATAI',///!IX,'REACTOR PARAMETERS',IX,
* 'CATALYST PARAMETERS',6X,'THERMAL WELL PARAMETERS',//
!IX, 'EPS •' ,E12.4,!1X, 'CPS •' ,E12.4,9X, 'CPT •' ,E12.4/
!IX, 'l •',E12.4,9X, 'PS •' ,EIZ.4,9X, 'PT •' ,E12.4,/
* !IX,'R• •',E12.4,9X,'TC •',E12.4,9X,'KT •',El2.4/
* !IX, 'RI •' ,E12.4/////6X,
* 'HEAT TRANSFER PARAMETERS',l9X,'REACTOR PARAMETERS'//
* !IX,'OHSG •',ElZ.4,!1X,'OHlA •',El2.4,!1X,'EXPA •',E12.4,/
9X,'OHTS •',Et2.4,!1X,'OHIB •',EIZ.4,!1X,'EXPB •',El2.4,/
•* !IX,'OHTG •',E12.4,9X,'OH2A •',E12.4,9X,'EXPC •',ElZ.4,/
!IX,'BGS •',E12.4,9X,'DH28 •',EI2.4,9X,'KP2A •',£12.4,/
9X,'BSG •',ElZ.4,9X,'K6 •',!!Z.4,9X,'KPZ8 •',ElZ.4,/
!IX,'BTS •',E12.4,!1X,'KBP •',EIZ.4,9X,'K2 •',E12.4,/
!IX,'BTG •',E12.4,36X,'K3 •',E12.4/
•* !IX,'PTZ •',E12.4,36X,'EA1 •',E1Z.4/
* 9X,'PTR •',E12.4,36X,'EA2 •',El2.41
.a FORMAT 11111, 3)(, 'PROGRAM CONDITIONS 1 • II. 9X,
* 'TIME STEP •',FB.4,/9X,'MAXIMUM TIME •',F8,,,/9X,
'DISTURBANCE LENGTH •',F15.4,/9X,
* 'RADIAL COLLOCATION POINT •',F8.4,/9X,
'START TIME tHRSl •',F15.4,/9X,
* 'ACCURACY OF CONVERGENCE •',E1 •• 3,/9X,
'HAX VALUE OF DT •',F8.4l
5 FORMATI////,3X,'STEADY STATE CONDITIONSa',/I/,9X,'XCH4 •',
• El2.A,9X,'T• •',El2.4,/9X,'XCO •',E12~4,9X,'TW
E12.4,/9X,'XC02 •',El2.4,9X,'PT. •',E12.~,/9X,'XHZ •',
•',
*
* E12.4,9X,'PT1 •',£12.4,/9X,'XH20 •',E12.4,9X,'MG •'
* EIZ.4,/!IX,'UG. •',E12.4,9X,'CPG •',E12.4l
FORMATC514/5F8.2)
' END

Table A5·2 Continued

 -460-

SUBROUTINE OUT~UT (V,U,T,MMJ

C THIS ROUTINE OUTPUTS THE CALCULATED RESULTS.
IMPLICIT REAL*& IA-H,O-Zl
REAL•& L,MG
DIMENSION VllBBJ,UI8l ·
. COMMON /COMAT/ NE,N!5l,H!5l,ZII6l,NSI6l,Zt5,2Sl,AI5,ZS,2Sl,

..
815,25,25)
/OPCON/ SCH4,SCO,SC02,SHZ,SH20,PT.,PTl,STg,srv
/RADIAl WIIBI,WPI9,2l,OETAI2l,RR
.
* /GASPA/ CPG,PTZ,PTR,MG,PGS,UGS,UM
/REACP/ EPS,L,RB.RI
• /INDEX/ NSE,NTS,NTT,NTG,NCO,NC2,NUG,NNE,NE1
DATA RG /B2 .• 544Dg/
IFIMM .EQ, 11 PRINT I
IFIMM .EO. 2l PRINT 6
IFIMM .EO. 3l PRINT 8
PH2•PT1/PTg
PHI•U/Ill
TA•T*L/UGS
IF!MM .EO. •>PRINT 7,TA
PRINT Z
DO IBB K•l,NE
CALL INDICESIK,NK,Nl,N2,N3,N4,N5,N6l
PRINT 3, TRIV!NTG+Kl,VINTT•KI,2,PHll,VIMTG+Kl,
* TRIVINTG+KI,VtMTT+Kl,2,1.DBl,TiltVINTS+Kl,VtNTT+KJ,l,PHil,
* VINTS+Kl,TRIVINTS+Kl,VINTT+Kl,l,l.DBJ
t•• * DO lBB l•l,NK
PRINT 3, TRIYIN3+!J,VIN2+1l,2,PHIJ,VIN3+1l,TRIVIN3+1l,VIN2+Il,Z,
l.DIIl,TRIVINI+Il,VINZ+Il,l,PHll,YINI+Il,TRIVINl+Il,YIN2+Il,
* 1.1.0•>
PRINT 3, TRtVIHTG+NEil,VIHTT+NEll,2,PH!l,YINTG+NEll,
TlttVINTG+NEil,VINTT+NEll,Z,I.DBl,TRIVINTS+NEil,VINTT+NEll,
* l,PHil,YINTS+NEll,TRIVINTS+NEil,YINTT+NEII,I,l.DBl
PRINT 4
DO zgg K•l,NE
CALL IHDICESIK,NK,NI,N2,N3,N4,NS,N6l
CALL ACTCONCIYINCO+Kl,VINC2+Kl,Xl,X2l
PRINT 5, VINTT+Kl,XI,XZ,VIHUG+Kl
DO 'ZB/11 I•I,NK
CALL ACTCONCIYIN4+ll,VINS+Il,Xl,X2l
zr• PRINT 5,VIN2+I),XI,X2,YtH6+Il
CALL ACTCOHCtVtMCO+NEil,VINCZ+NEll,Xl,X2l
PRINT S,VINTT+NEll,XI,XZ,VINUG+NEil
RETURN
1 FORMATIIHI,//ZX,'IMITIAL GUESSES• ',//)
2 FORMAT116X,'GAS TEMPERATURE',ZSX,
• 'SOLID TEMPERATURE',//8X,'R • Rg',7X,'It • RR',7X,
'It • R1'.9X,'R • Rg',7X,'R • RR',7X,'R • Rl'l
3
* FORMAT15X,31F9.5,4Xl,ZX,31F9.5,4Xll
4 FORHATI/19X,'WELL TEMP.',SX,'CO CONC.',5X,'C02 COMC.',4X,
5
• 'VELOC lTV' /l
FORMAT<19X,F9.5,5X,F1B.7,3X,F1 •• 7,3X,F9.&,3K,F9.Sl
FORMATI//2X,'STEADV STATE SOLUTION• ',//)

•'
7 FORMATI//,2X,'TIH£ M 0 ,F11.4,' SEC'/)
FORHATI//,2X,'ACTUAL STEADY STATE SOLUTION• ',//l
END

Table A5-2 Continued

 -461-

SUBROUTINE RADIAL
C THIS SUBROUTINE CALCULATES THE CONSTANTS ~OR THE RAOIALLV lUMP£0
C MODEL.
IMPLICIT REAL*B tA-H,O-Zl
COMMON /DIMLE/ ALS,ALG,ALT,BES,BEG,GAS,GAG,GTS,GTG,DEI,
* OE2,Sll,Sl2,SI3,PHI2,PHI3,PH1,PH2,PH3
/RADIAl WllBl,WPI,,Zl,DETAIZl,RR
• /HEATT/ OHSQ,OHTS,OHT,,BGS,BSQ,BTS,BTG,BWG.BWS
Vt!ll••. D•
Wti•>••.D•
1•1
SI•8TS
S2•81JS
S3•GAS
S.t•BES
SS•GTS
t•• * DETAtll•tPHI+PH1**2*tRR-1.D.l+RR**2*11.D.-PHll-RRl*Sl*S2
+ t2.DB*IPHI-RRI+RR**2-PHI**2l*SI
* + t-l.OB+RR**2+2.DB*tPHl-PHI*RRll*S2+2,DB*PHl-2.DB
WPtl,ll•tPHl*SI-l.DBl*IS2+2.0Bl-lS2+1.DBl*tPH1**2*Sl-2.D.*PHll
WPt2,ll•SI*tS2+1.DBl*RR**2-SI*RR*tS2+2.DBl
WPt3,Il•IS2*RR*tPH1**2*SI~2.DB*PHll-S2*1PH1*SI-I.D.l*RR**2l
WPI4,ll•S2*1PHI**2*SI-2.DB*PHil-Sl*IS2+2.DBl
WPIS,ll•Sl*IS2+2.0Bl-SI*S2*RR**2
WP16,1l•IS!*S2*RR**2-S2*1PHI**2*SI-2.DB*PHill
WP17,ll•SI*IS2+1.DB>-S2*1PHI*SI-1.06l
WPI8,ll•RR*SI*S2-SI*IS2+1.0Bl
WPI9,ll•IS2*1PHI*SI-l.DBl-RR*SI*S2l
W!1+1*3-31•14.DB*WPt7,ll+WP!4,ll/RRl*S4/DETA!ll-S3
Wt2+I*3-3l•I4.DB*WPI8,ll+WPIS,ll/RRl*S4/DETAtil
W(3+1*3-3l•!4.DB*WP19,ll+WPt6,ll/RRl*S4/DETA!Il
W17+l-li•S5"1WPII,li+WP14,li"PHI+WP17,11*PHI••zl/OETA11l
Wl9l•W!9l+SS*!WPI2,ll+WPIS,ll*PHI+VPt8,ll*PHI**21/DETA!Il-SS
WIIBl•WIIBl+SS*IVPt3,ll+WPt6,ll*PHI+VPt9,Il*PHI**2l/DETAIIl
IF! l .EO. 2lRETURN
1•2
!:1 •8TC:
S2•BWG
S3•GAG
S4•BEG
SS•I;:TG
100 TO 1. .
END

Table A5-2 Continued

 -462-

SUBROUTINE FNI IY,Vl

C THIS SUBROUTINE DEFINES THE ALQEBRAIC EQUATIONS FOR USE BY NSES!
IMPLICIT REAL*B IA-H,O-Zl
REAL *8 MG
DIMENSION VllBBl,VIll
COMMON /COMAT/ NE,N15l,HI5l,ZII6l,NS16l,ZI5,Z5l,AI5,Z5,Z5l,
* 815,25,25)
/OIHLE/ ALS,ALG,ALT,BES,BEG,CAS,GAG,GTS,GTG.DEl.
* DE2,Sil,Sl2,SI3,PH12,PHI3,PH1,PHZ,PH3
* /LINEA/ AAI75,75l,B8175,Bl,DDI75l,UI8l
* /RAOIA/ WI1Bl,WPI9,2l,DETA12l,RR
* /HEATT/ OHSG,OHTS,OHTG,BGS,BSG,BTS,BTG,BWG,BVS
* /INDEX/ NSE,NTS,NTT,NTG,NCO,NC2,NUG,NNE,NE1
/iASPA/ C~G.PTZ,PTR,MG,PGS,UGS,UH
DO !BAr K•l,NE
CALL INDICESIK,NK,Nl,N2,N3,N4,N5,N6>
DO lAJI I•I,NK
PT•PRESII,Kl
CALL REAC IVIN4+Il,VIN5+1l,VINI+Il,VIN2+1l,PT,R1P,R2Pl
VII+Nll • ALS*SUMIV,I,K,1,2l + Wlll*VINl+Il + Wl2l*VINZ+l)
•* +
+
GAS*V!N3+1l + DEI*RIP*Il.OAJ+PHIZ*VINI+Ill
DE2*R2P*I!.OB+PHI3*VINI+Ill + W13l*PH3
Vll+NZl • ALT*SUMIV,I,K,2,2l + W17l*VIN1+1l + Wl8l*VIN3+Il
* + W!9l*VIN2+Il + W!1Bl*PH3
VII+N3l • -VIN6+ll*SUMIV,I,K,3,1l + IALG*SUMIV,I,K,3,2l
* + W14l*VIH3+1l + W!5l*VIN2+Il + GAG*V!Nl+Il
* + W16l*PH3l/PT*YIN3+Il
VII•N4l • -VIN6+ll*SUHIV,I,K,4,1l + SI2*VIN3+ll*R2P/PT
* - Sll*VIN3+ll*RIP/PT
VII+N5l • -VIN6+Il*SUMIV,I,K,5,1l - SI3*VIN3+Il*RZP/PT
VII+N6l • VIN3+Il*SUMIV,I,K,6,1l + VIN3+ll*VIN6+Il*
* IPH2-1.0Bl/PT- VIN6+1l*SUMIY,I,K,3,1l
1•• CONTINUE
00 3BAJ L•1,6
N7•6*NSE+IL-ll*NE1
DO 3BII K•Z,NE
3gg VIN7+Kl • SUMIY,NIK-Il+l,K-t,L,tl - SUMIY •• ,K,L,tl
VINTS+ll • SUMIY,.,I,l,ll - BGS*IVINTS+ll-YINTG+Ill
VINTS+NEll • SUMIV,NNE,NE,l,l~- BGS*IYINTG+NEil-YINTS+NEill
VINTT+Il • VINTT+Il • Ul3l
V<NTT+NEll • SUMIY,NNE,NE,2,1l
VIHTG+ll • SUMIV,.,1,3,1l + BSG*IVINTS+ll•YINTC+lll
-11.011-1.011/YINTG+lll/ALG
VINTG+NEll • SUMIY,NNE,NE,3,1l- BSG*IV!NTS+NEll•VINTG+NElll
VIHCO+Il • YIHCO+ll-UI4l
VINCO+NEll • SUMIY,NNE,NE,4,1l
VINC2+1l • YINCZ+Il-UI5l
VINCZ+NEll • SUMIY,NNE,NE,5,1l
VINUG+ll • VINUG+ll-UI2l
VINUG+NEll • YINTG+NE1l*SUMIV,NNE,NE,6,1l + YINUG+NEtl•
• VINTG+NEll*IPH2-l.DgltPH2- YINUG+NEll*
• SUMIY,NNE.NE,3,1l
RETURN
END

Table A5~2 Continued

 -463-

SUBROUTINE CPCALCISHZ,SCO,SCOZ,SHZO,SCH.,ST.,MG,CPG,CPG1,CPG2l
C THIS SUBROUTINE CALCS THE HEAT CAPACITY
IMPLICIT REAL*BIA-H,O-Zl
REAL*S MG
SH2•l.DB-SH2-SCO-SC02-SH20-SCH4
MG•SCH4*16 •• 430B+SC0*28.BlB40B+SCOZ*44.BS98D.+SH2*2 •• 16D•
* +SH20*18.BI54DB+SN2*2B.B134DB
CPGl•ISHZ*.BID-3+SNZ*.77570-3+SC0*1.2D-3+SCOZ*4.216D-3
• +SH20*7.17D-3+SCH4*12.4282D-3l/MG
CPG2•1SHZ*6.6ZDS+SN2*6.77SOB+SC0*6.60B+SC02*8 ••5126D.
+SHZO*ll.ZDB•SCH4*S.ZBB8DBI/HG
CPG•CPGl*STS+CPGZ
RETURN
END

FUNCTION SUMIY,I,K,L,Ml
IMPLICIT REAL*B IA-H,O-Zl
DIMENSION YIIBBl
COMMON /COHAT/ NE,NISl,HISl,Zil6l,NSI6l,ZIS,Z5l,AI5,25,251,
• 815,25,25)
SUH•B.BDB
IPl•I+l
NN•(L-Il*NSINE+ll+NSIKl
N7•6*NSINE+II+IL-II*INE+ll
IFIM .EQ, Zl GO TO
DO IBB ll•l,NIIO
z••
t•• JPl•J+l
SUH•SUM+AIK,IPl,JPll*YIJ+NNl
SUH•SUM+AIK,IPl,ll*YIN7+Kl+AIK,IPl,NIKl+Zl*YIN7+K+ll
SUM•SUM/HIKl
RETURN
z•s DO ZBl J•I,NIKl
JPI•J+l
z•t SUM•SUH+BtK,IPI,JPll*YIJ+NNI
SUH•SUM+BIK,IPl,ll*VtN7+Kl+81K,IPl,NIKl+Zl*YIN7+K+ll
SUH•$UM/HIKI/H!Kl
RETURN
END

SUBROUTINE AEAC tY1,Y2,TS,TG,PT,RlP,R2Pl

C THIS SUBROUTINE CALCULATES THE DIMENSIONLESS RATES
IMPLICIT REAL*B IA-H,O-Zl
REAL*B MG,KPZA,KPZB,KCZB,KlP,KZP,KPZ
COMMON /OPCON/ SCH4,SCO,SCOZ,SHZ,SH20,PTB,PTl,STB,STW
• /REACZ/. EXPA,EXPB,EXPC,KCZB,KPZA,KPZB,KIP,KZP,EAI,EAZ
• /LINEA/ AAI75,75l,88175,8l,DDI75l,UI8l
GA•B.DB
TT•STB*TS
KP2•0EXPIKP2A+KP28/TTI
TH1•SCO*IUI41-Yl+Ut5l-Y2l
THZ•SCO*!UI5l-YZl
THJ•l.DS-Z.OB*THl
YHZ•tSCO*UI7l-3.0B*TH1-TH2l/SCO/TH3
YHZO•ISCO*UI6l+THl+TH2l/SCO/TH3
YCH4•1SCO*UIII•THll/SCO/TH3
YCO•Vl/TH3
YC02•Y21TH3
I'A•PT*TH3
lFIYHZ .LT. B.DBl YH2•1.D-t•
lf(yco .tt. B.d6l yco•l.d•16
RlP•DEXP!-EAl/TTl*PA**Iexpa+expbl*YH2**expb*YCO**expa/
• (I.OS+K1P*PA*YCO+KZP*PA*YH2l**expc
RZP•DEXPI-EA2/TTl*(.830B+.170B*PA•PT•>*IYC02*YH2-YCO*YH20/
• KPZ•KCZSl
liE TURN
£110

Table A5-2 Continued

 -464-

FUNCTION TRIT.TT,L,Rl
IMPLICIT REAL*8 IA·H,O-Zl
COMMON /RADIAl WIIBl,WPt9,2l,OETAI2l,RR
* /LINEA/ AAI75,751,8Bt75,81,0Dt75l,U18l
TR•IIWPII,Ll•WP14,Ll*R+WPI7,Ll*R**2)*T
* •IWPI2,L)+WPI5,Ll*R+WPI8,Ll*R**Zl*TT
* +WPI3,LI*UIIl•WPI6,LI*Uill*R+WP19,LI*Uil)*R**Zl/DETAILI
RETURN
END

FUNCTION RFN IAA,NIIl

IMPLICIT REAL*8 IA-H,O-ZI
DIMENSION AAII88l
ST•B.DB
DO lBJJ I•l,NN
188 IFIDABSIAAII)) .COE. ST> ST•AAtll
RFN•·DLOCOIBIDABSISTII
RETURN
END

SUBROUTINE ACTCONCIYI,Y2,Xl,X21
IMPLICIT REAL*9 IA-H,0-21
COMMON /OPCON/ SCH4,SCO,SC02,SH2,SH20,PTB,PT!,STB,STW
* /LINEA/ AAI75,751,8Bt76,81,DDI751,UI81
THI•l.DB·2.D8*1SCO*IUI,l•VI+UI5l•Y211
l<l•Vl*SCO/THI
l<Z•VZ*SCO/THI
RETURN
END

FUNCTION PRESII,kl
IMPLICIT REAL*B IA-H,O-Zl
COMMON /OPCON/ SCH4,SCO,SC02,SHZ,SH20,PTI,PT1,STB,STW
/COHAT/ NE,NI6l,HI5l,ZI16l,NSI61,ZIS.251,AIS.Z5.2SI,
815,25,251
PRES•IIPTI-PTBl*IZIIKl+HIKl*ZIK,I+lll+PTil/PTI
RETURN
END

SUBROUTINE INDICES tk,NK,Nl,N2,N3,N4,NS,N61

IMPLICIT REAL*B IA·H,O·ZI
COMMON /COMAT/ NE,NISl,HISI,ZII61,NSI6l,ZIS,251,AIS,Z5,251,
.. 815,25,251
IISE•NSIII£+11
IIK•NIKI
111•NSIKI
112•NSE+NS!Kl
113•2*NSE<NSIKI
N4•3*NSE•NSIKI
NS•4"NSE•NSIKI
N6•S*NSE+NS!Kl
RETURN
END

Table A5·2 Continued

 -465-

SUBROUTINE COLLOC
C THIS ROUTINE CALCULATES THE ZEROS OF THE ORTHOGONAL POLYNOMIAL AND
C SETS UP THE AXIAL COLLOCATION MATRICES A AND I
IMPLICIT REAL*8 IA-H,O-Zl
DIMENSION DIIZ5l,D21Z5l,D3125l,V11ZSI,V2125l,R1261
COMMON /COMAT/ NE,NISl,HCSl,ZII6l,NSI6l,ZIS,2SI,AI6,25,26l,
• 815,25,25)
/INDEX/ NSE,NTS,NTT,NTG,NCO,NC2,NUG,NNE.NEI
PRINT 1
DO Ul/1 K•l, NE
NK•N!Kl
CALL JCOBII25,NK,1,1,11.D/I,/I.DII,DI,DZ,D3,RI
PRI"T z, K,H!Kl,(~!Jl,J•I,NK+Zl
DO IIIII l•l,NK+2
CALL DFOPRIZS,NK,I,I,I,I,Dl,D2,D3,R,Vll
CALL DFOPRI2S,NK,!,l,I,2,DI,D2,03,R,V21
DO 1112 J•I,NK+2
AIK.l.JI•Vl!Jl
BIK,I,Jl•V21Jl
1112 CONTINUE
ZCK, I l•RI I l
1.. CONTINUE
NS!ll•/1
Zllll•/I.D/1
DO 21111 K•I,NE
NSIK+ll•NSIKl+NIKl
2111 ZIIK+Il•ZIIKl+HCKI
PRINT 3,111ZIIKl+HIKl*ZIK,JII,J•l,NIKI+11,K•l,NEI,l.DII
NEI•NE+l
NSE•NSINEll
NNE•NINEl+l
NTS•6*NSE
NTT•6*NSE+NE1
NTG•6*NSE+Z"NEI
NC0•6*NSE+3*NEI
NC2•6*NSE+ .. *NEI
NUG•6*NSE+S*NEI
RETURN
I FORHATIIH1,/,3X,'ORTHOGONAL COLLOCATION OF FINITE ELEMENT$')
2 FORMATI///2X,'ELEHENT ~',13,5X,'LENGTH • ',FS.3,//,7X,
" 'COLLOCATION POINTS•',//31111X,6FB.4/Il
3 FORMATI//1//ZX,'COLLOCATION POINT SUHHARYI',//,6116X,6F8.4/ll
END

SUBROUTINE INTLSS!Y,U,EPI
C THIS ROUTINE CALCULATES THE INITIAL STEADY STATE PROFILE
IMPLICIT REAL*8 IA-H,O-Zl
DIMENSION Yll/llll,UI81
COMMON /INDEX/ NSE,NTS,NTT,NTG,NCO,NC2,NUG,NNE,NE1
EXTERNAL FN I
CALL INITIAL IY,Ul
CALL OUTPUTIY,U,/1.0/1,11
CALL NSESII6*1NSE+NE1l,Y,EP,211/I/I,II,FN1l
CALL OUTPUTIY,U,/1,0/1,21
RETURN

END

Table A5-2 Continued

 -466-

APPENDIX6

I..IN1!J\R PACKED BED REACTOR IIODEL

linearization of Reaction Rates

Methanation

If we neglect the second and higher order terms in a Taylor series expansion, the

linearized rate is of the form

where

R:U. = [8Ry
oz l..
After performing the di.trerentiation
 -467-

R
»yeo
=[-1-+
Yeo
_£_[_1-+ _1-+ _1_+
1-D YH 2o YCH4 Yeo
JL_l
~
+ 1.5
YH2
 -468-

Steam-Shift

YCOeYH2 -
Kcs ]
Kps YcoYHaO

where

If we neglect the second and higher order terms in a Taylor series expansion, the

linearized rate is of the form

where

Rs. = [ aa; J..

Then after differentiation
 -469-

4x2o
1-26
l
Rs.
 -470-

linearization of Algebraic Equations

The boundary conditions for the mass balances and for the energy equation

for the thermal well can be solved explicitly for the concentrations and thermal

well temperatures at the axial boundary points as linear expressions of the con-

ditions at the interior collocation points. However. the set of four boundary

conditions for the gas and catalyst temperatures are coupled and are nonlinear

as a result of the convective term in the inlet boundary condition for the gas.

After a Taylor series linearization of this term around the steady state inlet gas

temperature, gas velocity, and inlet concentrations, the system of four equa-

tions is solved for the gas and catalyst temperatures at the boundary points:

t
i=l
AoJ9•J

tAN+tJe•1
j r:l

= s tAo.;esi + r
i=l

f: AN+lJe~t
=1

with

and

-1
(Ao.c-'Aga) Ao.N+l Xgu 0
AN+l.O (AN+l.N+l +l.gu) 0 -A,z.
s = Aug 0 (Ao.o-~-G) Ac.N+l
0 -A.zs ~+1,0 (AN +1.N +1 +A.ze)
 -471-

Linearization and solution of the continuity equation for the velocities at

the interior collocation points and at the end of the reactor are somewhat more

complex. Atter linearization as described in Section 3.6 and after substitution

of the linearized equations above for the endpoint temperatures, the continuity

equation was solved for the velocities:

where

The matrices E, F', and G have the following elements 1

fori,j =1. ... , N+l

For i,j =1, ... , N:

1. Note that ~ are the elements of the above S matrix.

 • 472-

Fori= N+l, j = 1, .... N:

linearization of the Ditferential Equations

The linear reaction rate expressions and the linear expressions for the velo-

cities and for the concentrations and temperatures at the axial boundary points

are then substitu~ed into the differential equations. Due to extreme coupling

between the state variables, lengthy manipulation is necessary to linearize the

resulting equations. The linearization is completely analytic and the resulting

 - 4'73-

model is2

i=Ax+lhr

where

The elements of the state and control matrices A and Bare as follows. 3

Elements of the State Matrix

For i,j =1, ... , N (i ¢j)

2. In terms of deviation variables.

3. Note that the At.j and Btj terms on the left of the equalities correspond to the elements of the
state and control matrices A and B, whereas those on the right of the equalities correspond to
Lag.r~~git~.n derivatives a::~ given by Equation (3.3-43). The tcrma JL..j correspond to clcmenta of
the above H matrix.
 .. 474-

A.:.2N+j =0 A.:.2N+i =~

A.:.3N+j =0

~.4N+j =0

AN'+i.2N+j =0

AN+i,BN+.i = AN+i.::JN+i = ~+i.4N+j = ~H.4N+i = 0

 ~ 475-

~N+iJ =0 Ac!N+i,\ = Co>?

~N+t:,N+j =0

A_
.M.2Nt1,2N+j : lXt
[o.
J.J\j -
~.N+l~+l.j ]
AN+l.N+l

~+t:.3N+i : ~N+\.3N+i = ~N+i.4N+j = ~N+i.4Nt1 = 0

A_ _ u.
.n.:~N+i,j - -•.iiJ
[~[A.
2.,; .nor;,t -
AN+u:~.NH ]-
Yeo
+A.
'"""·o -
AN+1.oAi.N+t
k"'l AN+l.N+l ., AN+l.N+l

A_ - u
.n.:~N+t:,t - -·"'·"
[ J1, [A.
2.,; '"""..t -
Al=l
~+t..t:At.NH
AN+l.N+l
]-
Yeo.,
+A.
nf.,O -
ANH.ol\.Ntt
AN+l.N+l
l

.n.:~N+i,N+J 1... [A""'.t _

,._ . . = -H·l.N+J·[ JY, AN+I.tAt..N+l ]-
A.. Yeo.,
+A.
""'·o
_ AN'+t.oAt..N+l
A
]
~=1 r>.N+l.N+l N+l.N+l

A- - H\.N+i [ Jl.
.n.:~N+i.N+i - -
[A.
~ '""".A:
A:=l
-
AN+l..AI~.N+l
AN+l.N+l
A.
Yeo + .n.t.o
]-
.,
-
AN+t.oAi.N+l
AN+l.N+l
l

AaN+i,2N+j : AsN+i,2N+i : 0
 -476-

_ [
AsN+i.,3N+j ::: -Vi( t\J - AN+t.;At.N+l
A ••
.nt'4 +l.N +1
l

AsN+i..4N+i ::: 0

A _
E>4N+i.,j - -
H
i,j
[ Ji. [A.
~ "".A: -
AN+t..e.A.t.N+I ]-
Yco2
A. yoC0
+ 0>\,0 ~+l.OAt.N+l
A
-o
Y~
]
11.. 2 -
.t=l r>fll+l.N+l i: N+t.N+l

A _ l-L
C>4N+i..-,; - -·'1- ...:
lr ~ [A
~ n..t..e -
AN+t . .tAt .N+l ]-
Y~
+ n..t.oYcaa
A. -o -
~+1.oAt..N+t -o ]
Yeo2
k=l AN+l.N+l lc AN+l.N+l

A
"'lN+i.,N+; - -
_ H
!..N+i
[ Ji. [A.
~ ""'L.k -
AN+t.A:At.N+t ]-
A... Yco 2
+ ""'.oYco
A. -o AN+1.oAt..N+t -o ]
Ycaa
2 - A...
k=l .nt'4+l.N+l lc .nt'4+1.N+l

~N+i.,N+i.
~
=-l-4.N+i. [ .t=t [
~ At...e -
~... uAt..N+l ]-
Y~ + At.oY~AJe -
-~ AN+t,oAt.N+l
Y8aa ]
~+l.N+l 1: AN+I.N+l

~N+i.,2N+j ::: ~N+i.,2N+1. ::: 0

-ei(
A.,N+i. 3N+i.
' = --aS -R....
PT --.,c~
 -477-

A -
A4N+\,4N+j - - !A.
- V&( r>i,j - ~+lj~,N+l l
--'---=--''---
AN+I.N+l

where

Elements of Ccmtrol Matri:z:

To define the clements of the matrix ~ the following definitions are needed. For

the methanation system.

where after substituting for the concentration of nitrogen in terms of the other
chemical species .
 -478-

Based on readily available data for the beat capacities of the gases,

ale 0.000776 IX20 6.775

ala2 0.000034 IX2~ -0.155

a teo 0.000424 IX:! co -0.175

a leoI! 0.003440 1X2eos 1.276

alaeo 0.006394 1X2~o 4.425

alCB4 O.Ol16b3 ~CB4 -1.574

Then let us define

 -479-

XCo(ToCltHaO + ael\!0)
a.= ~Cp.

Then if we define

the original definition for the constant r becomes

The matrix I can then be partitioned using the definition of vector E

where fori = 1, ... , N

 -480-

and fori = N+l

Thus we get

Also if we define
 -481-

Then the elements of the control matrix B are

 -482-

1:1_ _ 1:1. a.,B,;.N+l~+UJ

'B:aN+i.2 =0 .I.J"lN+i,3 - a.,..L.I\,0 -
ANH.N+l

'B:aN+i,4 =:&.!N+i,5 =B2N+i,6 ='B:aN+i,7 ='B:aN+i,B =B3N+i,l =0

BsN+t.2 =-l~[At.o - ~+l,oA,,N+1

AN+l.Nt 1
+ t
A:=l
(A,J: - A,,N+l~+lJ:
AN+t.N+l
)Yco11 ]
 -483-

- - a.,, e...... A.N 1AN a

+ CXe z 1(
T::L
.u;jN+\.4 =--"'i.ovii
A.
a1
... R
P,. u:10 + a2-p
Rs -
""""·A..+ +1. -
vi(
co, T Y&, rl.N+l.N+l

B4N+i.l =0

B4N+\.2 -- I [A.
- 2
'
-0
-"'i.OYCOa - ~+t.oAi.N+l
AN+I.N+l
.9
Yeo 2
+ ~
l.J (A.
~=I
-"'i,A: - At.N+lAN+l):: )-
AN+l,N+l
Yeo
eA:
I
B _
4N+\.3 - - 14 0
lr A_ -o
-"'i.oYco11 -
AN+1.o~.N+1
..4
•>-N+l.N+l
uo
JC0 2
+ Jl.
l.J (A_
A:=l
.t"1.\.A: -
~.N+l~+l.A:
A..
l"l.N+l.N+l
)-
Yeo"'·
"'
I
+ a1 z2,
 ~ 484-

B4N"'-"" _ A;. - + At.N+lAN+l.O _ -o

V8,yrn_ -
EJ8( R
CJ3 - p S !L
+ CX3 Z 2,
T..... - - ·oVa,
. ... A..
~+l.N+l
• ""'1:
T Yw~
•
 -485-

REFERENCES
 -486-

Ahmed, N.U., and K.L. Teo, Optimal Control of Di.stritruted Parameter Systems.

North Holland, New York (1981).

Aidarous. S.E .. M.R. Gevers and M.J. Installe, Int. J. Control. 22, 197-212 (1975).

Alvarez, J .. J.A. Romagnoli and G. Stephanopoulos, Chern. FJng. Sci., 36, 1695-

1712 (1981).

Amouroux. M.. G. DiPillo and L. Grippo, Richerche di.Autom.atica, 7, 92 (1976).

Ascher, U., J. Christiansen and RD. Russell. Math. Camp .. 33, 659-679 (1979).

Balas, M.J., Int. J. Control, 29, 3, 523-538 (1979).

Beek, J., Advances in Chemical .Engi:neeri:ng, Vol. 3, Academic Press, New York

(1962).

Belletrutti, J., andA.G.J. MacFarlane, .Proc. lEE. 118. 1291-1297 (1971).

Betts, J.T., and S.J. Citron. AIAA Jrru:rnal, 10, 1, 19-23 (1972).

Bird, R.B., W.E. Stewart and E.N. Lightfoot, 'Jlra:nsport Phenom.l!'na, John Wiley &

Sons, NewYork (1960).

Bonvin, D.. "Dynamic Modeling and Control Structures for a Tubular Autothermal

Reactor at an Unstable State," Ph.D. Thesis, University of California, Santa

Barbara. CA (1980).

Bonvin, D.. R.G. Rinker and D.A. Mellichamp, 'Tubular Autothermal Reactors:

Dynamic Modeling in the Region of Multiple Steady States," presented at

AIChE Meeting. San Francisco (1979).

Bonvin, D., R.G. Rinker and D.A. Mellichamp, Chem. FJng. Sci., 35, 603-812 (1980).

Bradshaw, A., and}3. Porter, Int. J. Control. 16. 2, 277-285 (1972).

Bristol, E.H.. Jnstrum.entatinn in the Chemical and Petroleum Industries. 16.

151-158 (1980).
 -487-

Brockett, RW., Pi:nite Di:m.ensional Linear Systems, John Wiley & Sons, New York

(1970).

Brown, K.M., Comm. ACM, 10, 11, 728-729 ( 1967).

Bryson, A.E., and Y-C Ho, Applied Optimal Control, Blaisdell, Waltham, Mass.

(1969).

Carberry, J.J., Ol.emical and Catalytic Reactor Engineeri:ng, McGraw-Hill, New

York (1976).

Carberry, J.J., and J.B. Butt, Ca.tal. R8V., 10, 221 (1975).

Carberry, J.J., and M. Wendel. AIOI.E J., 9, 129-133 (1963).

Carberry, J.J., and D. White, Ind. /ic Eng (hem, 61. 27-35 (1969).

Carey, G.F., and B.A. Finlayson, Ol.em. Eng. Sci., 30, 587-596 (1975).

Chidambara, M.R., and E.J. Davison, IEEE 'Prans. Auto. Control, AC-12, 119-121

(1967a).

Chidambara, M.R., and E.J. Davison, IEEE 'Prans. Auto. Control, AC-12, 213-215

(1967b).

Chidambara, M.R., and E.J. D~vison, IEEE 'Prans. Auto. Control, AC-12, 799-800

(1967c).

Clement, K., and S.B. Jorgensen, 8th IFAC World Congress, Kyoto, Japan, Aug 24-

28 {1981).

Clement, K., S.B. Jorgensen and J.P. Sorensen, Chem. Jf}ng. Sci., 35, 1231-1236

(1980).

Dahlquist, G., and- A. Bjerke, translated by N. Anderson, Nu:m.erical Analysis,

Prentice-Hall, Englewood Cliffs, New Jersey (1974).

 -488-

Davis, M.E., Num.erical Methods a.nd. Mod.el?:ng for Ol.em:iaa.l Engi:neers, John Wiley

& Sons, New York (1984) ..

Davison, E.J., IEEE Trans. Auto. Cantrol, AC-11, 93-101 (1966).

Davison, E.J., IEEE Tra.ns. Auto. Control, AC-13, 214-215 (1968).

Davison, E.J., IEEE Tra.n.s. Auto. Control, AC-15, 348-350 (1970).

Davison, E.J., and KJ. Chadha, Automatica, 8, 263-273 (1972).

De Wasch, A.P., and G.F. Froment, Ol.em. FJng. Sci., 26, 5, 629-634 (1971).

Douglas, J.M., Process Dyna.mics a.nd. Cantrol, Vol. 2, Ccmtrol System Synthesis,

Prentice-Ilall, Englewood Clit!s, New Jersey (1972}.

Edmunds, J., and B. Kouvaritakis, Int. J. Control, 29, 5, 787-796 (1979).

Ellis, J.K., and G.W.T. White, Control, 9, 82, 193-197 (1965a).

Ellis, J.K, and G.W.T. White, Control, 9, 83, 252-256 (1965b).

Ellis, J.K. and G.W.T. White, Control, 9, 84, 317-321 (1965c).

Fabien, R.W., and l.M. Stankovic, Ch.em. FJng. Sci., 34, 1350 (1979).

Fan, L.T., K.C. Chen and L.E. Erickson, Ol.em. FJng. Sci., 26, 379-387 (1971).

Ferguson, N.B., and B.A. Finlayson, Gi.em. FJng. J., 1. 327-335 (1970).

Ferguson, N.B., and B.A Finlayson, A!Ol.E J., 18, 1053-1059 (1972).

Finlayson, B.A., Chem. /!}ng. Sci., 26, 1081-1091 (1971).

Finlayson, B.A .. The Method. of Weighted. Residuals a:nd Variational Pri:nciples,

Academic Press. New York (1972).

Finlayson, B.A .. Ca.t. Rev. Sci. FJngng., 10, 1. 69-138 (1974).

Finlayson, B.A., Nonli'near Analysis in Chemical .FJngi:rwfffing, McGraw-Hill, New

York (1980).
 -489-

Fisher, R.J., and M.M. Denn, A!Ot.E J., 24-, 3, 519-523 (1978).

Foss, A.S., AJChE J., 19, 2, 209-214 (1973).

Foss, A.S., J.M. Edmunds and B. Kouvaritakis, I&EC Fundamentals, 19, 109-117

(1980).

Fossard, A., IEEE 'Prans. Auto. Control, AC-15, 261-262 (1970).

Froment, G.F., Adv. Chern. Sf!T'., 109. 1-34 (1972).

Froment, G.F., Chem. /ng. Tech., 46, 9, 381 (1974).

Gould, L.A., Ot.emical Process Control: Theory a:nd. Applications, Addison Wesley,

Reading. Massachusetts (1969).

Gould, L., and M.A. Murray-Lasso, IEEE 'Prans. Auto. Control, AC-11, 4, 729-737

( 1966).

Grad, J., and M.A. Brebner, Comm. ACM, 11,12 (1968).

Hallager. L.. and S.B. Jorgensen. 8th IFAC World Congress, Kyoto, Japan. Aug 24-

28 (1981).

Hansen, K.W., and S.B. Jorgensen, Adv. Otem. Sf!T'., 133, 505 (1974).

Hansen, K.W., and S.B. Jorgensen, Chem. Eng. Sci., 31, 579-566 (1976a).

Hansen. K.W .. and S.B. Jorgensen. Chem. E/ng. Sci., 31, 587-598 (1976b).

Hlavacek, V., Ind. FJng. Chem., 82, 7, 8-26 (1970).

Hlavacek, V., and M. Marek, Ot.em. Eng. Sci., 21, 501 (1966).

Heiberg, J.A., B.C. Lyche and A.S. Foss, AIOJ.E J., 17, 6, 1434-1447 (1971).

Hsu, C.H., and C.T. Chen, Proc. IEEE, 56, 206_1-2062 (1968).

Jutan, A., J.P. Tremblay, J.F. MacGregor and J.D. Wright, AIChE J., 23, 5, 732-742

(1977).
 -490-

Karanth, N.G., and R. Hughes, Cltt Rev Sci Eng, 9, 169 (1974a).

Karanth, N.G., and R Hughes, Ol.em. Eng. Sci., 29, 197-205 (1974b).

Khanna, R., and J.H. Seinfeld, Int. J. Control, 36, 1. 1-24 (1982).

Kouvaritakis, B., Int. J. Control, ?:1, 5, 705-724 (1978).

Kouvaritakis, B., and A.G.J. MacFarlane, Int. J. Ccmtrol, 23, 2, 149-186 (1976a).

Kouvaritakis, B., and A.G.J. MacFarlane, Int. J. Ccmtrol, 23, 2, 167-181 (1976b).

Kouvaritakis, B., W. Murray and A.G.J. MacFarlane, Int. J. Ccmtrol, 29, 2, 325-358
(1979).

Kouvaritakis, B., and U. Shaked, Int. J. Control, 23, 3, 297-340 (1976).

Kumar, S., and J.H. Seinfeld, IEEE Trans. Auto. Control, AC-Z:J, 690 (1978a).

Kumar, S., and J.H. Seinfeld, Ol.em. Elng. Sci., 33, 11. 1507-1516 (1978b).

Leden, B., IEEE Trans. Auto. Control, AC-21, 3, 408-410 (1976).

Lee, A.L., ''Methanation for Coal Gasification," Sym.p. on Clean Fuels from Coal,

Chicago, Sept. 1973.

Lee, E.B., and L. Marcus, Found.ations of Optimal Control, Wiley, New York
(1967).

Lynn, L.L., and R.L. Zahradnik, Int. J. Control, 12, 1079-1087 (1970).

MacFarlane, A.G.J., Autama.tica, 8, 455-492 (1972).

MacFarlane, A.G.J., and J.J. Belletrutti, Automa.tica, 9, 575-588 (1973).

MacFarlane, A.G.J., and B. Kouvaritakis, Int. J. Control, 25, 6, 837-874 (1977).

MacFarlane, A.G.J., _and I. Postlethwaite, Int. J. Control, 25, 1. 81-127 (1977).

MacGregor, J.F., and A.L.K. Wong, '1dentification and Stochastic Control or a

Packed Bed Reactor," presented at AIChE Symposium, Miami (1978).

 -491-

Mahapatra, G.B., IEEE Trans. Auto. Control, AC-:?2, 3, 481-482 (1977).

Marshall, S.A., Control, 10, 642-643 (1966).

Matsumoto, J., and K. Ito, I'Tii. J. Control, 12, 3, 401-419 (1970).

McCausland, I., Proc. lEE, 112, 3, 543-548 (1970).

McGlothin, G.E.. Int. J. Control, 20, 3, 417-432 (1974).

McGoWin, C.R., and D.D. Perlmutter, Ot.em. I!Jng. J., 2, 125-131 (1971).

Mears, D.E., Ind. J!)ng. Ol.em. Pundam., 15, 1, 20-23 (1976).

Michelsen, M.L., H.B. Vakil and A.S. Foss, l&EC Pundame'Tiials, 12, 3, 323-328

(1973).

Moe, J.M., CEP. 68, 3, 33-36 (1962).

Newman, C.P., and A. Sen, l'Tii. J. Co'Tiirol, 16,539-548 (1972).

Norrie, D.H., and G. DeVries, The Pi:nite Element Method, Academic Press, New

York (1973).

Ogunnaike. BA. and W.H. Ray, A!Ol.E J.. 25. 6, 1043-1057 (1979).

Paris, J.R., and W.F. Stevens, The Canadian J011.rna.l of Chemical Flngi:neeri:ng, 48,

2 (1970).

Pearson. C., J. Math. Phys., 47, 351-358 (1968).

Perry, R.H., and C.H. Chilton, Ol.emical .FJngineers ·Handbook, McGraw-Hill, New

York (1973).

Porter, B., and A. Bradshaw, Int. J. Control, 15, 4, 673-681 (1972).

Prabhu, S.S., and I. McCausland, Proc. lEE, 117, 7, 1398-1403 (1970).

Rabinowitz, P., ed., Numerical Methods for Nonlinear Algebraic Equations, Gor-

don & Breach, New York (1970).

 -492-

Ray, W.H., Advanced Process Control, McGraw-Hill, New York (1981).

Rosenbrock, H.H., CEP. 58, 9, 43-50 (1962).

Rosenbrock, H.H., .Proc. lEE. 116, 11, 1929-1936 (1969).

Russel, R.D., and J. Christiansen, SIAM J. Num.. Anal., 15, 59-80 (1976).

Sakawa, Y., IEEE Trans. Auto. Control, AC-9, 4, 420-426 (1964).

Schwarl:t, C.E., and J.M. Smith, Ind. Bng. Chem.., 46, 1209 (1953).

Seborg, D.E., and T.F. Edgar. eds., Ol.em.ical Process Control 2, United Engineer-

ing Trustees, New York (1962).

Shaw, I. D., "Modeling and Discrimination Studies in a Catalytic Fluidized Bed

Reactor," Ph.D. Thesis, McMaster University, Hamilton, Canada (1974).

Sheirah, M.A. and M.H. Hamza, Int. J. Control, 19. 5, 691-902 (1974).

Silva, J.M., "Multi-Bed Catalytic Reactor Control Systems: Configuration, Develop-

ment and Experimental Testing," Ph.D. Thesis, University of California,

Berkeley (1976).

Silva, J.M., H. Wallman and AS. Foss, I&EC FUndamentals, 18, 4, 363-391 (1979).

Sinai, J., andA.S. Foss, AJOLEJ., 16,4,,656-669 (1970).

Smith, J.M., Chemical.EJngine.eri:ng K'I:TI..etics, McGraw-Hill. New York (1970).

Sorensen, J.P., Oz.~m.. EJn.g. Sci., 31, 719-725 (1976).

Sorensen, J.P., Ol.em.. EJn.g. Sci., 32, 763-774 (1977).

Sorensen, J.P., S.B. Jorgensen and K. Clement, Ol.em. EJn.g. Sci., 35, 1223-1230

(1980).

Stewart, W.E., Ol.em.. FJng. Progr. Symp. Ser., 61. 56, 61 (1967).
 -493-

Stewart, W.E., and J.V. Villadsen, AIOJ.E J., 15, 28 (1969).

Strand, D.M., 'Design and Construction of a Fixed Bed, Non-Adiabatic Methanator

for Multivariable Computer Control Studies," Ph.D. Thesis, California Insti-

tute of Technology, Pasadena, CA (1984).

Vakil, H.B., M.L. Michelsen and A.S. Foss, l&EC F'u.nda.menta.ls, 12, 3, 328-335

(1973).

Valstar, J.M., P.J. Van Den Berg and J. Oyserman, Oa.em. EJng. Sai., 30, 7, 723-728

(1975).

Vatcha, S.R., 'l\nalysis and Design of Methanation Processes in the Production of

Substitute Natural Gas from Coal," Ph.D. Thesis, California Institute of

Technology, Pasadena, CA (1976).

Villadsen, J.V., and M.L. Michelsen, Solution of DiJ!ere:ni.ial Equation Models by

Polynomial Appro:z:imation, Prentice-Hall, Englewood Cl.ifis, New Jersey

(1978).

Villadsen, J.V., and W.E. Stewart, Oa.em. FJn.g. Sai., 22, 1483-1501 (1967).

Vortmeyer, D., K.J. Dietrich and K.O. Ring, Ol.emica.l Reaction EJngi:neering II.

588-599 (1974).

Vortmeyer, D., and R.J. Schaefer, Ol.B'm. Flng. Sai., 29, 485-491 (1974).

Wallman, H., J.M. Silva and A.S. Foss, !&EC F'unda.mentals, 18, 4 (1979).

Wang, P.K.C., IEEE Trans. Auto. Control., AC-17, 4, 552-553 (1972).

Wang, P.K.C .. Siam. J. Control, 13, 2, 274-293 (1975).

Weast, R.C., ed., CRC Handbook of OI.B'mistry and. Physics, CRC Press, Cleveland,

Ohio (1976).
 -494-

Wilson, R.G., D.G. Fisher and D.E. Seborg, AIChE J., 20, 6, 1131-1140 (1974).

Wright, J.D., and L. Schryer, '1\n. Application of Multivariable Model Reference

Adaptive Control to a Pilot Scale Catalytic Reactor," presented at AIChE

Symposium. Miami (1978}.

Young, L.C., and B.A Finlayson, I&EC F'u.ndamentai.s, 12. 4, 412-422 (1973).