Cover Story

Distillation, Part 1·
Experimental Validation
of Column Simulations
Apractical look at the need for validation, as well as conceptual considerations and a case study
is article addresses

11
Glenn Graham,
Pratik Pednekar he need for experi-
and Don Bunning mental validation of
MATRIC computer-generated
simulations of distillation col-
umns, and discusses the main
IN BRIEF concepts involved with experi-
mental validation. Also included
DIFACULTlES is a case study of a simulation
ASSOCIATED WITH
effort that initially yielded some
SIMUlATlONS
uncertainties. These uncertain-
CONCEPTS FOR ties were readily resolved with
EXPERIMENTAL batch laboratory testing.
VALIDATION
A CASE STUDY Difficulties close enough to the conditions of the source
Potential problems with physical prop- data to allow the data to be confidently uti-
erty models. Process simulation tools, lized. If not, literature data may be available at
such as Aspen Plus, CHEMCAD, and Pro- the conditions of interest to allow regression
Sim Plus, are indispensable for the design and estimation of new property parameters.
and optimization of separation schemes There are numerous thermodynamic
involving distillation columns. However, the equations that can be used to predict the
accuracy of their predictions depends highly vapor-liquid equilibrium (VLE) or liquid-liquid
on the physical property models used in the equilibrium {LLE) between various compo-
simulation, and it is not unusual to obtain nents. The differences in the various ther-
significantly different simulation results when modynamic equations may be subtle and
using different physical property models. the selection of the most appropriate equa-
Some level of experimental testing is often tion may not be obvious. The thermody-
necessary to validate the simulation results. namic equation selection is often based on
The physical property models consist of rule-of-thumb techniques. Other times, the
pure component properties, thermodynamic simulator default equation can be used, or
equations and component-interaction pa- the equation may be obtained from similar
rameters used within the thermodynamic simulation work in the literature, which may
equations. Accurate values for the pure not have been validated. The interaction pa-
component properties and the interaction rameters between the various components
parameters, and the correct thermodynamic that go into the thermodynamic equation
equation are seldom known at the beginning are often not available in the simulation soft-
of a simulation effort. ware for all of the binary pairs of chemicals.
Some software packages offer built-in These interaction parameters often must be
pure-component-property libraries contain- obtained from literature or estimated using
ing data for various components. In these group contribution methods like UNIFAC
libraries, the parameters - such as those in (universal quasichemical functional group
the Antoine equation - are regressed from activity coefficients) or developed from new
built-in experimental data. The user must de- experimental testing.
termine if the conditions of the simulation are Knowledgeable and experienced engi-
30 CHEMICAL ENGINEERING WWW.CHEMENGONUNE.COM FEBRUARY 2018
neers can generally make good decisions liquid composition versus theoretical stage number
in the selection of the thermodynamic equa-
tion and the determination of the interaction 0.9

parameters when compiling the physical 0.8

property models. Nevertheless, uncertainty c
0 0.7
in the reliability of the simulation results often u... 0.6
remains. It is critical that these difficulties .:::
C/1

and uncertainties are properly considered ...E

C/1 0.5

in the development of the physical property '0 0.4

·:;
r:r
model to avoid distillation columns that fail to ::i 0.3
accomplish the required separation, or col- 0.2
umns that are overdesigned and incur un- 0.1
necessary costs. Lack of information about
0
the process chemicals (especially in the case 0 2 4 6 8 10 12 14 16 18 20
of new chemicals) and their interactions lead Theoretical stage number (condenser Is 1)
to uncertainty in the reliability of the pre-
dictions. Experimental distillation testing is onstrating the need for experimental validation FIGURE 1. Relatively small
recommended, either to verify the quality of simulation work is given later in the article). amounts of intermediate·
boiling compounds can build
of the simulation results, or to provide data Experimental testing may only be required for up to large concentrations
to allow the physical property models to be part of the separation scheme. A considerable within a distillation column,
calibrated to the subject process conditions. amount of judgement is necessary to decide interfering with an otherwise
Difficulties not captured by simulations. when experimental testing is appropriate and easy separation. Feed to stage
10 is 1.0 kg/ll of 50/50 melha-
Unexpected column chemistry may also what level of testing is needed. Often, espe- nol-heptanol. Feed to reboller
occur, generating troublesome byproducts. cially for early work, adequate experiments (stage 20) is 1.0 gill butanoL
These byproducts may cause product-qual- can be conducted in a simple batch column. Pressure is 760 mm Hg and
the reflux ratio is 2
ity problems or build up in the separation In other situations, a continuous column is
system, negatively affecting the process in needed. Later-stage work often requires
other ways. If the byproduct is an interme- equipment representing the entire process. A
diate-boiling compound, it may concentrate secondary decision is required regarding the
in the middle of the column, and could sig- scale of the equipment. The "scale" is gener-
nificantly affect the known separations within ally thought of as laboratory-scale equipment
the column. An example of this "bulge" ef- or a pilot -scale facility, but "lab-scale" and
fect is shown in Figure 1. The main feed to "pilot-scale" can cover a wide range of sizes,
the column in this example is methanol and which might overlap.
heptanol, which can be easily separated. If,
however, a small amount of butanol, repre- Concepts for experimental validation
senting an intermediate-boiling byproduct, is In any attempt to compare experimental dis-
fed to the reboiler, the butanol accumulates tillation results to simulation predictions, there
in the middle of the column and its concen- are three main concepts that interact to influ-
tration builds up to very high levels within the Condenser
column. This accumulation of an intermedi-
ate-boiling species would cause significant
Cooling medium
changes to the separation within the column
and would dramatically affect the tempera-
ture profile and the temperature sensor(s) \Reflux liqtJd Distillate product
used for control.
Other problems that cannot be predicted
by simulation include foaming and fouling.
Foam generation can reduce the capacity of _,...... Column (tray ()( pad<ed)
Feed miXture
a distillation column. In severe cases, foam-
ing will make the column completely non-
operational. Fouling of oontactor surfaces Stripping zone
may reduce the column's capacity and could
affect the efficiency of the contactor. Reboil vapor
The case for experimental validation. I FIGURE 2. Shown
These uncertainties associated with the physi- here Is a schematic
cal properties, as well as the potential for the Heating med•um
diagram of a con-
other issues that simulations do not address, ventional distillation
column with a teed,
dictate that empirical testing of distillation and distillation and
simulations be considered (a case study dem- Bottoms product ("residue•) bottoms products

CHEMICAL ENGINEERING WWW.CHEMENGONLINE.COM FEBRUARY 2018 31

 Cover Story
FIGURE 3. A positive error in McCabe-Thiele diag. for water-acetic acid at 760 mmHg VLE diagram for water-acetic acid at 760 mmHg
the predicted VLE curve of a
pinched binary system can Vl.E dlagram representing an error in the VlE diagram represenUng an error In the A~-7·
cause the required number 0·98 physical property model 0.9 physical property model ,.//.;
of stages to be significantly
underestimated (The graph at ... 0.96 Actual VlE diagram ... 0.8 ~../..&·~ ·
the right is an expanded view
i
G>
0.94 ~ 0.7 ActualVlE~m .•h/
of graph at the left)
g 0.92
~-··j
....,----.,...
;I:
c
0 0.6
:p :p
u u
~ 0.90 ~ 0.5
G> G>
~ 0.88 0 04
E
...
8. 0.86 8. 0.3
~ >"'
0.84 o.2
0,82 0.1

0.80
0.8 0.82 0.84 0.86 0.88 0.90 0.92 0.94 0.96 0.98 1.00 00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Liquid mole fraction water liquid mole fraction water

ence the comparison. These are as follows: complex form. Ukewise, ideas presented
• VLE of the components (and possibly LLE) using a conventional continuous column can
• Relative liquid-to-vapor flowrates (LJV be extrapolated to more complex columns
ratios within the column) with multiple feeds, sidestreams and so on.
• Vapor-liquid contactor efficiency (tray Vapor-liquid equilibrium. The relative volatil-
efficiency or packing height equivalent to ity of an ideal binary chemical system is equal
the theoretical plate, or HETP) to the ratio of the pure component vapor
When a disagreement exists between pressures. If the components' interactions are
experimental distillation data and the corre- not far from ideal and the two components
sponding simulation results, it will likely be a have sufficient relative volatility (>2), then small
difficult challenge to determine which one, or errors in the VLE predictions may not signifi-
more, of these effects are responsible for the cantly affect the separation. However, if the
discrepancy. Some additional detail is justi- relative volatility is small (<1.5), any deviations
fied to explain their interaction. between predicted and actual VLE will lead
For the sake of simplicity, the following to large differences between the actual and
discussion will refer to a binary system of predicted number of stages required to ac-
components and a conventional continuous complish the desired separation.
column with an overhead distillate, a bot- Non-ideal systems will often have a pinch
toms product, and a feed that is introduced at the upper or lower end of the binary VLE
somewhere within the column, as is shown curve. The term "pinch" means that the vapor
in Figure 2. Although real systems are sel- and liquid compositions of the light-boiling
dom binary, the relevant concepts are more component are nearly equal (this concept
readily explained using binary components. also applies to the heavy-boiling compound).
These same concepts will apply to multi- The pinch is usually at the upper (right-hand)
component systems in a similar, but more end of the VLE curve, which is the case for
McCabe-Thiele dlag. for water-acet.lc acid at 760 mmHg McCabe-Thiele dlag. for water-acetic acid at 760 mmHg
I 1
feedrate: 1 Kglh feedrate: 1 Kglh
09 Refluxratlo: 3 0.9 Reflux ratio: 3
Stages: 30 Stages: 30
0.8 RetlOiler duty: 1,028 kcallh o.eReboiler duty: 2,070 kcallh
Column heat loss: 0 kcallh Column heat loss: 1,000 kcallh
... ,!l
.!!! 0.7 Acetic acid In dlstlllate: 2.1 mol% 07
Acetic acid in distillate: 0.4 mol% Distillate
; Water in tails: 19.3 mOl% comp. ; · Water in tails: 4.6 mol% comp.
0
c 0.6 g 0.6
FIGURE 4. The left figure =u
represents a column with tl
~ 0.5 ~ 0.5
no heat loss, whereas in the G> .!!
0
right figure, 50% of heat from E 0.4 ~ 0.4
reboiler Is lost through the ... ...0
column walls. Laboratory 8. 0.3 ~ 0.3 feed line
columns with many trays at ~ >
high temperature could easily 0.2 0.2
have 50% heat loss or greater.
0.1 0.1
The increased Internal reflux Tailscomp.
improves separation, but also 0
increases reboiler duty and 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 I
reduces column capacity Liquid mole fraction water liquid mole fraction water

32 CHEMICAL ENGINEERING WWW.CHEMENGONUNE.COM FEBRUARY 2018

binary pairs that repel each other. In this small changes in the reflux ratio can cause
situation, the actual vapor pressure of the significant changes in the separation. A
heavy-boiling component is nearly equal small error in the VLE prediction by the
to that of the light-boiling component, simulator could have a similar effect to a
due to the increased liquid-phase activ- small change in the reflux ratio. It may be
ity coefficient of the heavy-boiling com- difficult to determine which is causing any
ponent at infinite dilution. With sufficient discrepancies between a simulation pre-
non-ideality, the activity coefficient of the diction and the experimental results. If the
heavy-boiling component will become components are so non-ideal as to cause
even greater, causing the heavy-boiling phase separation within the column, then
component's actual vapor pressure to accurate LLE predictions will be neces-
exceed that of the light-boiling com- sary to produce accurate VLE predictions.
ponent at infinite dilution of the heavy- Liquid/vapor ratios. The relative vapor-
boiling component. This is the case for to-liquid flowrates within the column are
minimum-boiling azeotropes, and at the not only determined by the feed rate, reflux
azeotrope composition, the actual mix- ratio, boilup ratio, and distillate-to-bot-
ture vapor pressures (pure component toms ratio, but also by heat effects. Ex-
vapor pressure times activity coefficient cessive heat loss from a high-temperature
at the component's liquid concentration) column with many stages can significantly
of the two components will be equal. increase the internal reflux. This heat loss
The closer the composition is to a reduces both the liquid and vapor flow-
pinch zone or azeotrope, the greater will rates as one progresses up the column.
be the effect of any errors in the predic- The LN ratio will be the lowest at the top
tion of the VLE on the separation. Figure of the column, where it can be experi-
3 shows how a positive error in the VLE mentally determined from the reflux and
diagram can result in a significant under- distillate flowrates. However, the LN ratio
estimation of the number of stages re- will progressively increase as one moves
quired. Also, near a pinch or azeotrope, down the column, due to condensation of

TURNKEY FRACTIONAL DISTILLATION SYSTEMS

BATCH AND CONTINUOUS MODES
FOR OPTIMIZED PROCESSING,
FRACTIONAL DISTILLATION SYSTEMS
FROM POPE SCIENTIFIC OFFER THE WIDEST
RANGE OF CHOICES AND CAPABILITY:
• LAB. PILOT OR PRODUCTION SIZES, 0.1·1000 KG/ HR
• STAINLESS STEEL, HASTELLOY. GLASS, OTHER MATERIALS
• RANDOM, STRUCTURED, OTHER PACKING & INTERNALS
• MODULAR COLUMNS, 1" - 24" DIAMETER
• AMBIENT, VACUUM OR PRESSURE CAPABILITY
• OPERATION VIA PLC/COMPUTER PACKAGES OR BASIC CONTROLS
• TURNKEY SYSTEMS OR INDIVIDUAL KEY COMPONENTS OFFERED
• EXPLOSION PROOF, ASME, CE. OTHER RATINGS AVAILABLE

These proven systems are desiened for critical

applications includin& hi&h purit y specialty chemicals
and solvents, pharmaceutical intermediates, flavors,
f ragrances , foods, oils, bio-based materials, and more.

POPE'SDISTILLATION EXPERTISE · SEPARATING OUR

CLIENTS FROM THEIR COMPETITION FOR OVER 50 YEARS!

www.popeinc.com
Botch Mode System Cot~tlnuous Mode System
1-262-268-9300
For d e tails visit a dllnks.ch e m engonllne.com/70303·23
 Cover Story
FIGURE5. The Murphree McCabe-Thiele diagram with Murphree eHiclency concept McCabe-Thiele diagram with Murphree eHiclency concept
vapor efficiency represents
how close the actual vapor Actual equilibrium line
compositlon comes to reach- 0.9
ing equilibrium with the
liquid on an actual distillation
..
s
0.9
y·
n ..
~
0.8
Pseudo· equilibrium
line representing
Murphree effiCiency
tray. The concept of the non- "':&:c 0.8 :&:
c
0.7
attainment of equilibrium can 0 0 0.6
£""... 0.7 ...
:::>
be represented as a pseudo-
equilibrium line on a Mccabe-
Q)
"'
.::
Q)
0.5
Operating lines
Thiele diagram, in which 0.4
actual trays are drawn Instead
of theoretical trays [5)
..
0
E 0.6
0
a.
Murphree efficiency
equatl011: ..
0
E
0
a.
0.3
Fffil tine
"'
> 0.5 EMV= Yn. Yn.r "'
> 0.2
r;. Y•., 0.1
0.4 0
u u ~ v ~ u ,. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Liquid mole fraction water Liquid mole fraction water

the vapor. This effect can be thought of as the right diagram of Rgure 5. The McCabe-
an increase in the effective reflux ratio going Thiele concept could be applied to this
down the column. This increase in LN ratios pseudo-equilibrium curve, in which actual
due to heat loss will cause the column to trays would be drawn, instead of theoretical
have a better separation than predicted by stages, as for a true equilibrium curve.
a computer simulation, but at the expense of If the Murphree vapor efficiencies for the
higher energy costs and lower capacity than trays of a distillation column are 100%, the
expected (see Figure 4). Without understand- number of actual trays required to accomplish
ing that the actual LN ratios are different than a separation would be equal to the number of
expected, the VLE curve would appear to be theoretical trays required. When the Murphree
more open (higher relative volatility or less vapor efficiencies are less than 100%, the re-
pinched) than is actually the case. Preheated quired number of actual trays will be greater
feeds and reflux from the condenser that than this theoretical number. Therefore, if the
have heat qualities different than expected actual tray efficiencies are less than predicted,
can have similar effects on the column in- more actual trays will be required to accom-
ternal flows. Sub-cooled feed or reflux will plish a desired separation than would be an-
cause condensation, which increases the LN ticipated based on the required number of
ratio below its point of entry into the column. theoretical trays and the predicted efficiency.
Superheated or flashing feed may produce Without properly accounting for the reduced
vapor flowrates different than anticipated, af- efficiency, the separation will appear to be
fecting the LN ratio above the feed point. more difficult than predicted.
Efficiencies. Process simulators generally The efficiency of packed columns is rou-
use theoretical stages to represent the trays tinely expressed as HETP (height equivalent
in a distillation column. However, in real col- to a theoretical plate). If the actual HETP is
umns, the vapor leaving a tray is seldom in greater than predicted (less efficient), the
equilibrium with the liquid from that tray. The column will contain less theoretical stages
difference between the actual and equilib- than predicted and this will produce similar
rium vapor compositions can be expressed results to a column with trays that have re-
as an efficiency. A common method of de- duced efficiencies.
fining efficiencies for actual distillation trays The tray efficiency is dependent on sev-
is the Murphree vapor efficiency (see Figure eral factors, including the configuration of the
5). Values for Murphree vapor efficiency tend tray (hole size, number of holes, hole area) or
to be in the range of 50 to 75% for labo- packing (surface area, shape, void fraction),
ratory- and pilot-scale columns, but can lie the properties of the chemical mixture, the liq-
outside of this range. The Murphree vapor uid and vapor flowrates, the LN ratios, and
efficiency is an indication of how closely the the pressure. Several authors provide meth-
vapor composition in an actual distillation ods for estimating tray efficiencies [1-4].
tray approaches equilibrium with the liquid Effect of pressure on the separation. The
leaving that tray, as shown in the left-hand previous discussion shows how the VLE, LN
diagram of Figure 5. For a binary chemical ratio, and contactor efficiency are intertwined,
mixture, if the vapor composition for every and the importance of knowing the values of
actual tray were plotted against the liquid each of these for the experimental validation
composition on that tray, it would appear of distillation columns. Another variable that
as a pseudo-equilibrium curve, as shown in needs to be considered is pressure. Pressure

34 CHEMICAL ENGINEERING WWW.CHEMENGONUNE.COM FEBRUARY 2018

 Methanol-ethanol Vlf diagram at different pressures FIGURE 6. For the methanol-
not only affects the temperature, but can also
ethanol binary system, lower
change the VLE curve, as shown in Rgure 6. pressure opens up the VLE
0.9
Lower pressure often makes the VLE curve curve. which would make a
more open and the separation becomes eas- distillation separation less
ier at lower pressures.
Trayed plant columns usually have signifi-
-...
$
~
08

07
difficult

c
cantly higher pressure drops than laboratory .2 0 6
or pilot columns. Therefore, if a laboratory u...
~
o.s
or pilot column is used to validate simula- G> -76mmHg

-
0 04
tion results, and it is operated at the same E - 76mmHg
head pressure as that intended for the plant &. 0.3 760mmHg
column, the laboratory or pilot column will >"'
02
have a lower base pressure than the corre-
sponding plant column. For columns operat- 0.1

ing in the range from moderate vacuum to 0

above atmospheric pressure, the effect on 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Liquid mole fraction water
temperature and separation may not be sig-
nificant. However, for high-vacuum columns columns. Different physical property models,
with many trays, a laboratory or pilot column including NRTL (non-random two-liquid),
may have a much lower base pressure than UNIQUAC and Wilson, were tried based on
the commercial column and a correspond- the expected interaction of the components.
ingly lower base temperature. Generally, this The interaction parameters were estimated
lower pressure in the experimental column for the expected range of conditions for
allows easier separation properties than the each model. However, the simulation results
corresponding plant column. did not always demonstrate the expected
In addition to overestimating the separa- separation. The results did not appear cor-
tion capabilities of the plant column, the lab- rect and often were very different from one
boratory or pilot column may underestimate physical property model to another.
the amount of degradation that will occur in A simple batch distillation experiment was
the base of the plant column, since the plant set up to perform a validation experiment
column's base pressure and temperature will to confirm that the separation could be ac-
be higher than for the laboratory or pilot col- complished. The apparatus used was a 1-in.
umn. To overcome these problems, it may glass Oldershaw column with an evacuated
be necessary to operate the small-scale col- and silvered jacket. Aberg lass insulation was
umn at the same base pressure as intended added around the column and around the
for the plant column, or to run separate ex- top of the glass kettle. The starting mixture
periments, matching the head pressure in of the chemicals was heated in the base and
one test and the base pressure in another. the column was operated In total reflux until
it reached steady state. After that, the reflux
A case study ratio was changed and a total of ten distillate
An example of how simulation discrepancies samples were collected.
can be resolved using batch distillation ex- A software package called Aspen Batch
periments is presented here. Modeler was used to simulate the unsteady-
A client was interested in developing a state batch distillation. The software simu-
method to continuously separate a mixture lates the batch distillation process dynami-
of ethylene glycol (EG), propylene glycol (PG) cally, allowing the user to specify changing
and 1,2 butanediol (1 ,2 BOO) in water. Distil- experimental conditions, including reflux
lation was one of the options being consid- ratios, distillate collection and so on. This
ered for the continuous commercial sepa- software accesses the same Aspen proper-
ration. The components had similar boiling ties library as Aspen Plus, which was being
points and the desired purity could not be used to simulate the commercial distillation
achieved under modest positive pressures. columns. Thus, once the optimum physical
From past experimental knowledge and property model and parameters were found
experience, a subject-matter expert (SME) that allowed the batch simulation to match
was aware of how this separation would the batch experimental data, this model and
take place under vacuum, and the simula- the corresponding parameters could be
tion work progressed to consider vacuum used directly in Aspen Plus to simulate the
separations. Initial simulation studies evalu- commercial continuous case.
ated the separation of a known composi- The mass of the components in the over-
tion of the mixture using a combination of head distillate samples in the batch ex-

CHEMICAL ENGINEERING WWW.CHEMENGONLINE.COM FEBRUARY 2018

 Cover Story
FIGURE 7. The Wilson prop- NRTL property model used in the simulation to compare with Wilson property model used in the simulation to compare
erty model provides a much experimental results with experimental results
better fit than the NRTL model.
The above figures show the EG in overhead sample EG in overhead sample
results for the mass of ethyl-
ene glycol (EG) and propylene
glycol (PG) collected in the
i:
~4
"-J \
.. \
cos
~5
~ 4
overhead sample in the ex·
~ 3 ~3
periment as compared to the
~ 2 ~2
simulation. The figures on the
left compare the experimen- ! 1
;';o
~ I

tal data with the simulation ~ o L....,::::::::L~L--------- ..... 0 2 3 4 5 6 7 8 9 10

when using the NRTl property 0 2 3 4 5 6 7 8 9 10 Overhead sample number
Overhead sample number
model, while those on the - -eG exp - EG slm (NTRl) - . . EG exp - EG sim (Wilson)
right show a comparison
when using the Wilson prop- PG In overhead sample PG in overhead sample
erty model. The binary param- Ql 12
eters were estimated using
UNIFAC for both cases
t 10
..,~ 8
6

i.5
4
2
~ 00 2 3 4 5 6 7 8 9 10
2 3 4 5 6 7 8 9 10 Overhead sample number
Overhead sample number
- - PG exp- PG sim (N'TRl.) - - PG exp - EG sim (Wilson)

periment were compared to the simulation. Wilson case, however, compared much bet-
Because of the expected component interac- ter with the experimental results. Using the
tions, activity coefficient models were used. Wilson property model and optimized pa-
Specifically, the NRTL, UNIQUAC and Wilson rameters, the results for the mass fraction of
property models were individually evaluated. EG, PG and 1,2 BOO are compared with the
For each of these models, different param- experimental data in Figure 8. The data show
eter estimation techniques were tried. Binary a very good fit. The quality of this fit demon-
parameters were calculated by regressing strates that the Wilson property model cor-
data from the literature. The Antoine's param- responding parameters are appropriate for
eters were calculated from literature data at the subject conditions, and that they can be
the conditions of interest. Also, UNIFAC was used with confidence in the simulation of the
used to estimate the binary parameters at the continuous commercial columns.
operating conditions. The best match to the The commercial distillation column simula-
experimental data was obtained using the tions in Aspen Plus were then re-developed
Wilson property model. Figure 7 shows the utilizing the Wilson model and the data from
FIGURE 8. Mass fractions of close fit of the simulation results using the Wil- the experimental batch distillation. The revised
the components In the distil- son model and the contrasting poor fit when simulations showed distillation to be a feasible
late overhead samples are
compared with the optimized using the NRTL model. separation technology, including from a capi-
batch column simulation The results for the NRTL case showed tal and operational cost perspective. Later, a
results when the using the only EG and water (not shown in the Figure client SME validated the distillation column
Wilson property model. A very
good match was obtained
7) being recovered in the distillate. No PG re- simulation results based on experience in an
using the Wilson property covery was seen in the simulation in contrast existing plant with similar separations.
model to the experimental data. The results of the In this case study, the batch d istillation ex-
Simulal.ion versus e.xperimental composition of the disl.illate in the overhead samples periment was useful not only to validate the
0.9 feasibility of the separations, but it also pro-
- 0.8 vided data to calibrate the simulation model

I
~ 0.5
0.7
0.6
and to thereby add confidence to the com-
mercial predictions.

0
c 0.4 Concluding remarks
§ 0.3 The use of process simulation tools for mod-
£ eling distillation columns is invaluable for de-
~ 0.2
~ 0.1 • • signing plant-scale columns. However, it can
o r"-- ......---='-tr
be difficult to know if the simulator is gen-
1 2 3 4 56 7 8 9 1o erating accurate predictions. Additionally,
Overhead sample number there are potential problems associated with
- PGslm - - PGexp - EGsim - -EGexp - 1.28DOslm
•.2eoo exp distillation columns that process simulators

36 CHEMICAL ENGINEERING WWW.CHEMENGONUNE.COM FEBRUARY 2018

do not address. Hence, experimental distil- References
lation studies are quite important to either 1 S!ICiima~. J. G., aoo Far. J R.. 'Oislillation: Prirlo!Jies ood Pr~·
verify the simulation results or to provide a Jdvl Wlef & Sons. klc., Hoboken, N.J. 1998.
path forward to improve the simulations. In 2. Kister, H. Z. 'Oist.llaboo Oeslgn.' McGraw-ti11 Inc., New Vorl<, N.Y
1992
this article, a case study was presented of
3. Treybal, R. E., Mass·Transfe~ ()peraboos.' McGraw-Hl, Inc., New
a simulation effort that yielded some uncer- YOlk, NV., 1987
tainties, which were then readily resolved 4 Veri WI~. M , Disbllalion.'McGraw·Hil.lnc. New Vorl<, NY., 1967
with laboratory batch-column testing. 5 Kister H Z.,"DistlllationOeslgn,'McGraw·K , NewY0!1<, NY,1992.
When contemplating the need to experi-
mentally validate distillation simulations, key Authors
points include the following: ..------:=---.. Glenn Graham is a distillation SME and se-
• For process simulations, developing the nior chemical engineer at MATRIC (Mid-At-
lantic Technology, Research, & Innovation
proper physical property model by select- Center, P.O. Box 8396, South Charleston,
ing the correct pure component proper- wv 25303: Phone: 304-552-6554: Emall:
ties, thermodynamic equations and com- glenn.graham@matricresearch.com: Web-
site: www.matricinnovates.com). Prior to his
ponent interaction parameters can be a position at MATRIC, Graham worked for
challenging task, especially for moderate Union carbide Corp. and The Dow Chemical
Co. as a distillation specialist in their R&O
to severe non-ideal mixtures. separations groups. Graham holds a M.S.Ch.E. from Montana
• Without experimental validation, it is diffi- State University.
cult to know if the simulations are accurate Pratik Pednekar is a project manager and
• Byproduct generation, foaming and foul- research chemical engineer at MATRIC
(Email: pratik.pednekar@matricinnovates.
ing are difficult or impossible to predict by com). He works In the areas of process con-
process simulators. ceptualization and development. laboratory
• Experiments should be conducted to experimentation, pilot plant testing and
technology evaluation. Prior to working at
validate distillation simulations and to MATRIC, Pednekar was awarded his Ph.D.
determine if other non-simulated prob- from West Virginia University in the field of
lems will occur. process development, involving simulation.
optimization and reactor modeling.
• The main concepts that affect the separa-
Don Bunning is currently employed by MAT-
tion in a distillation column are the VLE, RIC (Phone: 304-720-1049, Email: don.
lN ratios, and contactor efficiencies bunning@matricresearch.com). He has ex-
(pressure also plays a role). tensiVe experience in process developmenl.
reaction engineering, new catalyst develop-
• When a disagreement exists between ment, demonstration of new technologies in
experimental distillation data and the cor- pilot units, commercialization. on-site startup
support, and licensing of the technologies.
responding simulation results, it will likely His more than 45 years of R&D experience in
be a difficult challenge to determine which ' - ' - - - - - - ' the chemical Industry includes technical and
one, or more, of these effects are respon- management positions at Union carbide and Dow Chemical. He
hold a M.S.Ch.E. from the University of West Virginia and is a
sible for the discrepancy. • registered P.E. in West Virginia.
Edited by Gerald Ondrey

Proven Vacuum Technology

for Chemical Processes
>Custom solutions to meet your requirements
>Safe pumping of hazardous flammable material
>Know-how from project conception to completion
>Worldwide service support

For de tails visit adlinks.c heme ngonline .c om/70 3 03· 07

CHEMICAL ENGINEERING WWW.CHEMENGONLINE.COM FEBRUARY 2018 37

Reproduced with permission of copyright owner. Further reproduction prohibited
without permission.