Basic Modelling of Dynamic Engineering Systems

Chapter 1
Basic Modelling of Dynamic

Engineering Systems
1.1 Introduction
This chapter1 presents an introduction to the basic issues of the modelling of dynamic
engineering systems. We will consider the role and use of models from a point of view that
is common in the engineering sciences.
The use of models has a long tradition in science. Yet, there is a difference in attitude
towards the use and role of models and modelling between the physicists and the engineers.
Consider for example the elementary properties of matter such as specific heat, viscosity,
or thermal conductivity. A physicist will not be entirely satisfied with these concepts
because they are not elementary enough. He will try to explain these concepts in terms of
the properties of more elementary concepts such as molecules, atoms, and electrons. The
engineer is also dissatisfied because these concepts in general are too elementary. For most
problems in engineering it is desirable to use these concepts merely as building blocks for
more elaborate concepts, by which the behaviour of larger processes can be explained. Thus
the engineer is interested in discovering the behaviour of macroscopic processes, whereas
the physicist will naturally concentrate upon the microscopic behaviour or will consider
nature at even smaller scale.
In engineering science, models constitute a main formalism for describing our knowledge
about engineering processes and systems. A model basically describes a limited number
of properties of the real world in terms of verbal, graphical, or mathematical language.
With the assistance of models we describe that part of the real world that is thought to be
relevant to the solution of a particular problem or to a class of related problems.
In the sequel, a systematic approach will be developed that enables to formulate dy-
namic mathematical models for physical systems and processes. Essentially, there exist
two different approaches for obtaining models.
1. Experimental modelling. The use of experimental data, obtained by measuring

and collecting the values of variables of the real system or process during well-planned
experiments, and through subsequent data processing, should discover systematic
relationships amongst the variables. These relationships are considered as a model.
We further may wish or require this model to be able to explain and predict the
behaviour of the studied system or process. The joint actions of experiment design,
1
sc4032 Physical Modelling for Systems and Control chapter 1
c O.H.Bosgra TUD 2009-2010
1
2 CHAPTER 1. BASIC MODELLING OF DYNAMIC ENGINEERING SYSTEMS
experimental data acquisition, data processing, and model validation are known under
the headings of experimental modelling, empirical modelling or system identification.
2. Theoretical modelling. Using the accepted theory of the underlying sciences, equa-
tions are derived that describe the basic behaviour of our system or process. In general
the equations involved will be conservation laws or other relations that have been de-
rived from first principles. Additionally, we need to describe the physical properties
and other relevant descriptions of the materials in the system or process. Thus, the
underlying theories that we accept to build the model are our basic axioms and as-
sumptions in the logical process of model construction. This approach will be called
theoretical or fundamental model building. It is also known under the name physical
modelling.
The steps to be followed in experimental model formulation in general require a considerable

amount of a priori knowledge. This a priori knowledge must be brought in a format such
that certain basic decisions regarding variables, experiments, and responses can be taken
with the required confidence. This implies that we must execute certain steps in the process
of theoretical modelling in order have a sufficient basis for experimental model formulation.
Thus the second approach can be considered as the natural starting point for any model
building activity. Additionally, in many cases the first approach is the natural sequel to
the second approach in the sense that it allows to verify or validate the structure and
parameters in a formulated model, or it may enhance and amplify certain insufficiently
understood or incompletely quantifiable parts of a fundamental model. In these notes an
approach to theoretical model formulation will be covered. In several cases the connections
with certain aspects of experimental modelling will be discussed.
The process of modelling is widely applied in many fields in science and engineering.
However, the amount of knowledge available for the formulation of theoretical models is
different in the various fields. In many fields of engineering there is sufficient knowledge
available to formulate simple over-all models that describe the behaviour of systems and
processes. However, for example in chemical process engineering the operation of most
chemical process equipment is so complicated that the detailed behaviour of chemical pro-
cesses can only in part be described accurately by mathematical models based upon our
present knowledge. Thus there are severe limits in our ability to formulate theoretical mod-
els for engineering systems. In many cases the use of experimental modelling techniques is
a requirement that by necessity complements the theoretical approach.
In these notes the formulation of models is aimed at obtaining a description of the
dynamic behaviour of processes under transient conditions. This implies that we will for-
mulate the equations of motion of the process variables that describe the evolution of the
process as a function of time. Our models will formulate the process dynamics in a form
as required for the understanding of process operations such as startup and shutdown, or
for studying the transitions from one operating condition to another one as, e.g., required
by grade changes in a production plant or by changes in the composition of the feedstock.
Process dynamic models also are of great importance for providing control engineers with
qualitative and quantitative descriptions of the transient behaviour of processes that are
to be used in model based control system design.
By its proper nature, a model is limited in the sense that the real world is approximated
by definition. Only those properties and phenomena that are relevant to the purpose of the
model should be included in the model. The real world is so highly detailed, and possesses
a rich complexity, that even a small part of the real world can never be exactly represented
in an isomorphic mapping sense by a single mathematical model, notwithstanding its high
1.2. THE FORMAL STEPS OF THE MODELLING PROCESS 3
number of equations and its great fidelity. Thus the process of model building is funda-
mentally a process of making appropriate and clever approximations. This fact forces the
modeller to consider the object of his activity from an over-all point of view and from a
certain distance. He must have an overview over the most important physical phenomena
in this object, evaluated from the point of view of the intended use or the foreseen purpose
of the model. The modeller must have an attitude that enables him to make decisions
regarding the most important phenomena to be embodied in the model. For this reason,
we must consider the process of modelling as a decision process.
A model can be defined as a “small representation of a planned or existing object”.
The adjective “small” connotes the central idea that a model is something less than real-
ity. Thus, in the sciences, models are sought which illuminate natural phenomena. The
objective is to strip away all phenomena that are not thought to be essential in view of the
intended use of the model. Thus a model must be able to explain those phenomenological
patterns of interest in terms of a set of easily understood elements.
In this context a model is a theory which constitutes a set of propositions or laws
from which may be deduced those facts which are exhibited in the real system or process
under consideration. More fundamentally, a theory must explain the less comprehensive
laws, established by observation of reality, in the sense of introducing ideas which are more
familiar or, in some way, more acceptable. Moreover, theory must predict new laws and
these laws must turn out to be true.
This last notion illustrates sharply the approach that has come to be called the scientific
method. The scientific method of establishing an understanding of any physical phenomena
is generally identified as consisting of the following three phases:
• Initial observation
• Formulation of a theory (in our language, a model)
• Experimentation (model (in)validation) and prediction of new observations (model

use)
The completion of the last stage will in many cases suggest refinements to the theory
(model), and the process is repeated. The emphasis on observations and experiments has
its roots in the empiricist philosophy which has been at the heart of modern science.
1.2 The Formal Steps of the Modelling Process

The scientific formalization of the approach to modelling sketched in the preceding section
can now be formulated and involves the following main features.
1. Defining the problem. The purpose of the model must be formulated and clarified.
This requires a certain mature familiarity of the modeller with the object to be
modelled. Also an understanding of the future model use will help in deciding which
properties of reality should be represented in the model. In general the modeller
must make decisions regarding the desired degree of detail in the model. Higher
granularity in general implies a higher amount of complexity, and in many cases the
dominant phenomena in the model can be represented with higher accuracy than the
minor details. Representing these details may be counterproductive in terms of the
intended use of the model. Thus unnecessary high detailing should be avoided.
z(t)
u(t) r internal y(t)
variables
m input output
variables system variables
surroundings
Figure 1.1: The system concept requiring r independent model equations
2. Define system boundaries and interconnecting variables. The process of

modelling is aimed at describing only a part of the real world. The system boundary
defines in its interior the system as opposed to its exterior which are the surroundings.
The modeller himself is part of the surroundings. The system and its surroundings are
coupled through signals which are process variables that originate in the surroundings
or in the system. Those variables that originate in the surroundings and influence the
system by exerting an action upon the system by passing across the system boundary
are called input variables contained in a vector u(t). We consider these influences in
a causal fashion, i.e., such that the cause precedes in time the effect. A system is
called causal if past and present internal and/or output variables do not depend on
future input values.
From a modelling point of view there is no distinction between the input variables that
can be manipulated at will, and those input variables that occur in the surroundings
as a result of unmodelled processes or of which the origin or cause is not exactly
known (disturbances). Finally, those variables that are the result of the relationships
within the system and which pass outwards across the system boundary such that they
become available to the surroundings are called the output variables of the system,
contained in the vector y(t). All variables that occur in the system and that are not
included in the set of input variables are called the internal variables, represented
by the vector z(t). See the representation in the figure 1.1. In general, the total
set of variables in a model setup is the sum of the input variables and the internal
variables. The output variables are to be considered as a subset of this total set of
variables. Note that the formulation of a mathematical model by its very nature
yields a set of variables that constitute the total set of variables mentioned before.
If this total set of variables is available as a result of the initial modelling decisions,
then it is a decision of the modeller to assume which subset of these variables is
specified by the surroundings and thus constitutes the set of input variables. The
remaining variables are the internal variables. As the input variables are specified by
the surroundings (i.e., by the modeller), these variables must be considered as known
variables in our problem setup. Then our model must be able to (uniquely) determine
the internal variables, given the input variables. This implies that in general we must
have a number of (independent) model relations that equals the number of internal
variables.
3. Development of a conceptual model. Once the model variables have been pre-
liminarily identified, the next step is the development of a conceptual physical model,
1.2. THE FORMAL STEPS OF THE MODELLING PROCESS 5
containing only those phenomena which are considered as relevant to the intended
use of the model. A thorough knowledge of the physics of the process plus intuition
gained from practical experience is valuable and important in this step. In its most
simple fashion, the conceptual model only consists in the mind of the modeller. The
model can be verbally described, and the relevance of incorporating or neglecting
certain phenomena can be discussed.
4. Formulation of model hypotheses. In this step the real physics in the process
are replaced by a simplified hypothetical physical representation. The precise as-
sumptions needed to arrive at this simplified representation are to be formulated in a
number of model hypotheses or model assumptions. In a certain sense the predictive
power of the resulting model is reduced by adapting its properties to the specific
situation at hand. This will be greatly counterbalanced in general by the increase in
model simplicity. The chosen degree of simplification determines the character of the
mathematical relationships that will finally represent the physical phenomena.
5. Decomposition into subsystems. In many cases, a modeller’s intuition can only

keep track of a limited number of phenomena simultaneously. For this reason it can
be beneficial to split up the complete process into a number of parts by formulating
spatial boundaries by which subsystems are created. The interconnection of all sub-
systems gives rise to the original system. The subsystems thus defined must satisfy
individually the requirements of a well-posed system. This implies amongst other
things, that the interconnection of the various subsystems should be consistent, the
number of internal variables should be equal to the number of (independent) model
relationships available for each subsystem, and the inputs and outputs of the original
system should be contained in the various inputs and outputs of the subsystems.
6. Formulation of conservation laws and other mathematical equations. The

hypotheses formulated in the previous steps must enable now to write down (if appro-
priate) the basic conservation laws for energy, momentum and mass. These equations
will be complemented with rate equations for mass and heat transfer, with descrip-
tions of the thermodynamic and other physical properties of the materials in the
process, and with all other fundamental or correlations-based relationships that are
needed in order to arrive at a physically and mathematically consistent model. The
result is a set of algebraic and differential equations having a certain mathematical
stucture, and containing a set of parameters that fix the model behaviour quantita-
tively.
7. Mathematical simplifications and approximations. The mathematical model

can be examined for its properties. Certain basic a priori requirements of a model
can be investigated, such as the properties of causality, continuity, and conditioning
with respect to possible numerical implementation for simulation purposes. Also
further properties such as stability, time scales, and independence of the set of model
equations can be established. On the basis of these findings it can be decided to
simplify or to approximate the mathematical representation of the system.
8. Numerical solution of the model equations, simulation. Further confidence

in the model can be gained by solving the model equations for well considered input
signals (simulation), and by observing the behaviour of the internal and output vari-
ables. By exercising with the model in this way, the model can inspire confidence or
the modeller can experience certain weaknesses of the model formulation.
reality
⇓ ⇐= intended model purpose
process under consideration
⇓ ⇐= select system boundary, inputs, outputs
system and surroundings defined
determine relevant phenomena
⇓ ⇐=
and formulate model hypotheses
conceptually simplified physical model
formulate conservation laws
⇓ ⇐=
and other model relations
mathematical model having
structure and parameters
⇓ ⇐= mathematical approximations
simplified mathematical model
⇓ ⇐= experimental model validation
model satisfies requirements ?
⇓ =⇒ yes, model ready for use
no, return to previous steps
Table 1.1: Basic decisions in the modelling procedure
9. Validation of the model and its hypotheses. Comparison of predicted model re-
sponses and similar responses obtained at the real system may provide the final proof
of confidence needed to validate a model. It must be stressed that the experiments
must be representative for the intended purpose of the model.
The main steps of the approach are listed in table 1.1.
1.3 Formulation of Conservation Laws for Process Models

The equations of conservation of some quantity (mass, momentum, energy) are fundamental
laws of nature. These equations apply both in the steady state and under unsteady state
conditions. We first will consider macroscopic conservation laws for mass and energy. These
are integral balance equations that are applied to a volume inside a macroscopic control
surface, as opposed to microscopic conservation laws that are formulated for an infinitesimal
or elemental volume.
1.3.1 Conservation of mass

The density and concentration of a chemical species that possibly reacts are often measured
in moles per unit volume. A mole of a species contains Avogadro’s number (6.02 × 1023 ) of
molecules. Moles can be converted to mass units by multiplying by the molar mass which
has units of kg/mole.
Consider a system defined by its (fixed) macroscopic control surface or system boundary S,
enclosing a (macroscopic) volume V as shown in figure 1.2. Stated verbally, a macroscopic
1.3. FORMULATION OF CONSERVATION LAWS FOR PROCESS MODELS 7
system boundary S
mass
φin ρAin
flow in
perfectly
mass mixed
φout ρA
flow out volume V
surroundings
Figure 1.2: System boundary for macroscopic mass balance
conservation law for mass for a chemical species A within the volume V reads:

d N et accumulation N et transport N et generation
= + (1.1)
dt of A in V of A through S of A in V
Note that the left-hand side of (1.1) specifies the rate of change of mass accumulation
within the control volume V , and not the amount of accumulated mass itself.
Let ρA denote the density of species A, then the accumulation term in the left-hand side
of the equation (1.1) describes the rate of change per unit time of the accumulated mass
of A within the volume V :

d
ρA dV (1.2)
dt V
This form of the left-hand side allows the density to have varying values over the macro-
scopic volume V . In fact the integral expression performs a summation over all elementary
volume parts dV of the contribution of each ρA dV to the total accumulation of mass of
A. This formulation in fact does not reflect the overall character of the macroscopic bal-
ance and is not practical. In terms of cause and effect, we must consider the right-hand
side of equation (1.1) as the cause, creating an effect in terms of a variation of overall
accumulated mass in the left-hand side. Our macroscopic conservation law determines the
overall accumulation and thus does not provide a mechanism to deal with spatially varying
accumulation phenomena. Consequently, the density ρA should be considered as having
the same value over the whole volume V . This can be achieved by assuming that there
exists a physical mechanism that creates so much interchange of material over the volume
V that ρA has virtually the same value over V . One possible assumption is that V behaves
as if it were perfectly mixed. Thus, under the assumption of V being perfectly mixed,
our macroscopic conservation law for mass reads
dρA
V = φk ρAk + G (1.3)
dt
k
system boundary S
h(t)
φi (t) φo (t)
Figure 1.3: Fluid buffering vessel
where the φk denote the volumetric flows across the boundary S each having density ρAk ,
and G denotes the generation of species A per time unit within V , e.g., as a result of
chemical reaction. In this expression, φk are positive and negative for in- and outgoing
flow, respectively. The k is supposed to run over all flows that cross the system boundary.
Note that under the assumption that the volume V behaves as a perfectly or ideally mixed
fluid, the density ρA has the same value over the whole volume V , whereas this value may
vary in time. The assumption of an ideally mixed fluid implies that it is assumed that each
element in the volume V will vary in time in exactly the same fashion. It also implies that
any outward flow across the boundary S has a density that equals ρA . Thus for one flow
in and one flow out, the macroscopic conservation law (1.3) for mass reads
dρA
V = φi ρAi − φo ρA + G (1.4)
dt
where now the negative sign of the outward-bound flow has been explicitly formulated in
the expression, and thus we consider positive values for the flows in the direction given by
the arrows in figure 1.3.
While the general principle of the conservation law (1.1) is maintained, the actual equa-
tions may have other forms than equation (1.3) or (1.4) depending upon the structure
of the system under consideration. Under the assumption that the density ρ of the fluid
is constant, we still can have variations of accumulated mass within the system, as the
following example shows.
1.3.2 Example: Mass accumulation in vessel

Consider the simple flow system shown in figure 1.3. The steps to be executed for modelling
this process are as follows.
1. Purpose of the model: To describe the behaviour in time of the fluid level h(t)
varying with the purpose of buffering variations in input mass flow
2. System boundary: Around the vessel
3. Variables across boundary: φi , φo , h. As there apparently is no mechanism
inside the system boundary to determine φi and φo , these variables are supposed to
be determined by the surroundings, i.e., φi , φo are inputs and h is output.
φi (t)
vessel h(t)
φo (t)
model
Figure 1.4: Block diagram of inputs and outputs of fluid buffering vessel
φi (t) initial h(0)
+
1
h(t)
− A
φo (t)
Figure 1.5: Internal structure of model for fluid buffering vessel
system boundary S
h(t)
X(t)
φi (t) φo (t)
pi (t) po (t)
Figure 1.6: Fluid buffering vessel with outward flow controlled by valve
h(t)
φi (t) vessel
+ pi (t)
X(t) valve
φo (t)
Figure 1.7: Inputs and outputs of fluid buffering vessel with outward flow determined by
valve
4. Model hypotheses: the fluid is incompressible; the vessel cross sectional area A is
constant; the velocity of the fluid inside the vessel is slow so that local variations in
fluid level can be ignored.
5. Conservation law: Mass balance

d
(ρAh(t)) = ρφi (t) − ρφo (t) (1.5)
dt
which can be written in differential equation (state space) form:
dh(t) 1 1
= φi (t) − φo (t), h(0) = h0 (1.6)
dt A A
Note that the model (1.6) can be depicted by the block diagram in figure 1.4 and
by the structural representation in figure 1.5. Also observe that the system is not
asymptotically stable, as the (open) integrator may give an unbounded response for
bounded inputs. This indicates, that the model represented by equation (1.6) is only
valid for h(t) ≥ 0 and h(t) ≤ hmax , where hmax indicates the height of the vessel.
A simulation of this process should include these bounds. If one of the bounds is
reached, the model (1.6) should be replaced by a different one, one for the empty
vessel case, and one for the overflow case.
6. Model satisfies requirements ? In retrospect, at this stage it is useful to re-

consider the assumptions and starting points of the modeling exercise, and to make
modifications if desired. For instance, if in reality the flow φo depends on the value h,
then one might wish to include this relationship in the model. Figure 1.6 represents
the situation where the outgoing flow is determined by the valve opening and by the
pressure difference across the valve. We assume that the fluid pressure behind the
valve po is atmospheric, as is the pressure above the fluid level. We assume further,
that the flow through the valve is turbulent, proportional to the square root of the
pressure drop across the valve. This pressure drop is proportional to the fluid level.
We have the following 5 variables: φi , φo , pi , X, h where X represents the input
command determining valve position. The model now consists of equation (1.6) and
the following two relations:

φo (t) = KAv [X(t)] pi (t) (1.7)
pi (t) = ρgh(t) (1.8)

where Av [X(t)] represents the valve cross-sectional area as a function of valve position
X(t), K is the valve coefficient, and g is the gravitational accelleration. Equation
φi (t) initial h(0)
dh(t)
+ h(t)
1 dt
− A
K

φ0 (t) pi (t)
∗ ρg
Av (t)
pi (t)
Av (X)
X(t)
Figure 1.8: Internal structure of model of vessel and valve
−3 valve area level

x 10
1.1
1.6 1
0.9
1.4
0.8
m2
1.2 0.7
0.6
1
0.5
0.8 0.4
0 5000 10000 15000 0 5000 10000 15000
−3 flow out valve inlet pressure

x 10
1.6 11000
10000
1.4
9000
1.2 8000
N/m2
m3/s
1 7000
6000
0.8
5000
0.6 4000
0 5000 10000 15000 0 5000 10000 15000
time (s) −−> time (s) −−>
Figure 1.9: Response of fluid buffering vessel on step in valve opening

−4 flow in level
x 10
12
1
10
0.8
m3/s
m
8
0.6
6 0.4
4 0.2
0 0.5 1 1.5 2 0 0.5 1 1.5 2
4 4
x 10 x 10
−4 flow out valve inlet pressure
x 10
12
10000
10
8000
N/m2
m3/s
8 6000
4000
6
2000
4
0 0.5 1 1.5 2 0 0.5 1 1.5 2
time (s) −−> 4
x 10 time (s) −−> 4
x 10
Figure 1.10: Response of fluid buffering vessel on step in inlet flow, valve opening fixed
−3 flow in
x 10
2
m3/s
0
0 5000 10000 15000
−3 flow out
x 10
2
m3/s
0
0 5000 10000 15000
level
1.5
1
m
0.5
0 5000 10000 15000

time (s)
Figure 1.11: Buffering vessel: response on stochastic inlet flow variations, valve opening
fixed
(1.7) represents the flow through the valve and equation (1.8) determines the actual
pressure drop across the valve. We have 5 internal variables and 3 model relations,
thus two variables should act as inputs, i.e., are assumed to be determined by the
surroundings of the process. As φo , pi , and h are determined by the model relations,
we can assume φi and X to be input variables. This is in accordance with our physical
imagination. The block diagram with input and output variables, and an internal
representation of the model structure are given in figures 1.7 and 1.8, respectively.
The following observations can be made:
• Comparing figure 1.5 with figure 1.8, we see that the representation of the con-
servation law represented in figure 1.5 is contained in figure 1.8. However, now
there is a feedback loop around the integrator.
• The sign of the feedback path is such that an increase in accumulation of mass
will increase the outflow of mass, i.e., there is negative feedback.
• Thus we may expect that the system will be asymptotically stabilized by the
feedback relationship. This implies that a permanent change in inflow φi or
valve position X(t) will lead to new equilibrium values for h(t), φo , and pi (t).
• The process is bilaterally coupled to the neighbouring flow system to the left
and to the right. This means that there is a mutual information exchange
between the process and the adjacent systems at both sides. The model expresses
this coupling by assuming the pressure at the right interconnection po to be
atmospheric, i.e., not to change in relation to the variables considered in the
process. Thus φo (t) can vary without causing a response in po (t). At the left
interconnection, φi (t) is assumed to be determined by the upstream flow process,
and our model determines the reaction in pi (t) which acts as an input variable
for the upstream process.
• In figure 1.8, the block indicated by (∗) represents a multiplication block, i.e., its
output is the multiplication of all (three) input variables. Similarly, the block
indicated by the square root sign determines its output as the square root of
its input. These blocks represent nonlinear signal operations and show that the
represented system is nonlinear, i.e., does not satisfy the superposition principle.
7. Numerical solution of the equations: The block diagram of figure 1.8 can directly
be used in a simulation tool for continous systems (e.g., Simnon or Simulink). The
following parameters have been used:
A = 1.0 m2 ρ = 980 kg/m3

g = 9.8 m/s2 K = 0.01
Av = 0.001 m2
∗ φ∗o = 0.001 m3 /s
where the asterix values indicate the used steady state. The resulting value for h∗
is 1/(0.98)2 = 1.0412. Figure 1.9 shows responses of flow out and fluid level on step
disturbances in the valve position. The nonlinear behaviour causes different responses
for the positive and negative step inputs. Figure 1.10 shows for fixed valve position the
response to inlet flow step disturbances. Figure 1.11 shows the buffering capabilities of
the vessel for fast stochastic variations in input flow, which are considerably reduced
in the outlet flow.
S
φ V
Ti (t)
φ
To (t)
T (t)
Figure 1.12: Process vessel under ideal mixing assumption
initial T (0)
dT (t)
Ti (t) + φ dt T (t)
− V
Figure 1.13: Internal structure of energy balance model for mixing vessel
1.3.3 Conservation laws for energy

The formulation of correct energy balances is not a trivial matter. See for example chapter
5 in Denn [13]. In general, the total energy of a system consists of the internal, kinetic and
potential energy terms. In many cases where the energy balance is dominated by thermal
effects, the kinetic and potential energy terms can be neglected. A simple energy balance
without generation phenomena for a one-phase incompressible fluid of constant density ρ
and constant specific heat Cp in a constant volume V reads
d
(V ρCp T (t)) = φk (t)ρCp Tk (t) (1.9)
dt
k
where V is the volume within a fixed control surface S, T (t) is the time-varying temper-
ature within the control surface, and Tk (t) are the temperatures of the volumetric flows
φk (t) through the surface S, having positive and negative values for inflow and outflow,
respectively. Note that the above conservation law is a macroscopic balance, i.e., we have
explicitly or implicitly assumed that all fluid particles inside V behave identically in time,
as if V were perfectly mixed.
1.3.4 Example: Energy accumulation in vessel

For the case of one flow in and one flow out, the process is shown in figure 1.12 and the
energy balance is
dT (t)
V ρCp = φρCp [Ti (t) − To (t)] (1.10)
dt
Note that this is a macroscopic overall balance equation, summing all heat balance con-
tributions in the total volume V . Due to this overall summation, each particle within V
1.4. MICROSCOPIC CONSERVATION LAWS 15
inlet temperature inlet temperature

51.2
100
51
90 0.02 width
50.8
degrees
degrees
80 50.6
70 50.4
50.2
60
50
50
49.8
0 1 2 3 4 5 0 1 2 3 4 5
outlet temperature outlet temperature

51.2 51.2
51 51
50.8 50.8
degrees
degrees
50.6 50.6
50.4 50.4
50.2 50.2
50 50
49.8 49.8
0 1 2 3 4 5 0 1 2 3 4 5
time (hrs) time (hrs)
Figure 1.14: Temperature response of mixing vessel on inlet temperature impulse and step
has temperature T (t), possibly varying in time but spatially uniform. This corresponds
to the assumption of an ideally mixed internal flow pattern, and we may replace To (t) by
T (t). The resulting energy balance can be written in state-space form, i.e., in a first-order
differential equation form with only a time derivative of a state variable in the left-hand
side:
dT (t) φ
= [Ti (t) − T (t)] (1.11)
dt V
The block diagram of this model is shown in figure 1.13. The effect of the feedback loop
shows a negative sign, which in this case implies that the system is asymptotically stable.
The equation is a linear first-order differential equation, having a time constant
V
τ= (1.12)
φ
which in physical terms equals the average residence time of particles flowing through the
vessel. The figure 1.14 shows the temperature responses for a vessel with time constant
(average residence time) τ = 1 hour.
1.4 Microscopic Conservation Laws

In the case that a concentration, density, temperature or any particles-related quantity
can not be assumed to be uniform over the volume V inside a control surface S, it is
not useful to formulate a macroscopic (integral) balance over the total volume V . In other
words, if a volume V cannot be assumed to behave like ideally mixed, one could resort to the
formulation of microscopic conservation laws. The formulation of a conservation law for the
elemental volume (δx, δy, δz) in V , and subsequently letting the δx, δy, δz approach zero,
yields a microscopic conservation law having the format of a partial differential equation.
For energy, a possible form is

2
∂T ∂T ∂T ∂T ∂ T ∂2T ∂2T
ρCp + vx + vy + vz =k + + +G (1.13)
∂t ∂x ∂y ∂z ∂x2 ∂y 2 ∂z 2
Here the first term represents the variation of accumulation of energy per time units, and the
next terms in the left hand side describe the net bulk transport of energy through the surface
S. The first three terms in the right hand side describe transport through the surface S by
thermal diffusion; and G denotes a net generation term. The temperature T (x, y, z, t) now
is a function of both time t and the spatial coordinates x, y, z. The parameter k represents
the thermal conductivity (kgm/s3 deg), and defines the thermal diffusion coefficient
k
DT := (1.14)
ρCp
where DT has units m2 /s. The three velocities vx , vy , vz represent the local bulk velocity
components of mass particles in the three coordinate directions. In deriving equation (1.13)
we have assumed that ρ, Cp and k are constant over V . As a preliminary observation, we
see:
• A microscopic balance such as 1.13 contains three different types of terms:
1. Variations per time unit of accumulated energy

2. Transport of energy through convective transport, i.e., as a consequence of
the fact that particles move with a velocity (vx , vy , vz ) whereas the coordinate
frame is fixed
3. Transport of energy through diffusion, i.e., through a physical mechanism of
molecular interaction that creates an energy flux proportional to (the negative
of) the actual spatial gradient of the temperature
• The flow pattern in a vessel must be known or must be computed in some way if one
wishes to formulate and to solve a microscopic conservation law
• For the derivation of mircroscopic balance equations such as 1.13, consult e.g., Him-
melblau and Bischoff [23] and Bird, Stewart and Lightfoot [9]. A derivation for the
one-dimensional case is given in the next chapter.
Along similar lines one obtains the following microscopic mass balance in three rectangular
spatial coordinates for a component A in a dilute binary system:
2
∂cA ∂cA ∂cA ∂cA ∂ cA ∂ 2 cA ∂ 2 cA

+ vx + vy + vz = DAB + + + GA (1.15)
∂t ∂x ∂y ∂z ∂x2 ∂y 2 ∂z 2
where cA is the molar concentration (mole/m3 ) of component A, and DAB (m2 /s) is the
diffusion coefficient.
Note that the microscopic balances (1.13) and (1.15) require the availability of the initial
values T (x, y, z, t0 ) and cA (x, y, z, t0 ), respectively, if we wish to determine their evolution-
ary solution in time for t ≥ t0 . These initial values require the specification of T and cA at
t = t0 over the whole spatial domain. Additionally, at the boundary of this spatial domain
the boundary values for T (x, y, z, t) and cA (x, y, z, t) and their partial space derivatives
have to be specified as a spatial function of time. Moreover, the formulation of the equa-
tions (1.13) and (1.15) assumes that the velocity field vx (x, y, z, t), vy (x, y, z, t), vz (x, y, z, t)
1.4. MICROSCOPIC CONSERVATION LAWS 17
x=0 x=L
T (x) T (x + ∆x)
v v
A
x x + ∆x
x
S S
Figure 1.15: Pipe flow with uniform velocity v(t) (plug flow)
is known as a function of the three spatial coordinates and time. These facts indicate
that in general a microscopic formulation either requires the parallel computation of the
actual possibly time-varying flow pattern, or requires considerable simplifying assumptions
in order to be of practical use. The first option can be realized by extensive numerical com-
putations using specialised numerical software packages, and will be utilised in cases where
a rigorous model is required. Simplifications may involve the observation that the flow pat-
tern can be idealized, e.g., in cases where the main flow direction is in only one coordinate
direction. In that event a one-dimensional spatial representation may suffice. Also note
that the occurrence of velocity and diffusion coefficients in the microscopic balance equa-
tions (1.13) or (1.15) implies that when these coefficients become large, then the behaviour
of the balance equations approaches the behaviour of the ideally mixed macroscopic case.
1.4.1 One-dimensional microscopic balance equations

We will present the derivation of microscopic balances for the one-dimensional case. Con-
sider the flow in a pipe having essentially a uniform velocity v(t) over its cross-sectional area
A. We call this type of (idealised) flow pattern plug flow. Applying a (macroscopic) en-
ergy balance to an elementary volume A∆x as shown in figure 1.15 and assuming constant
physical material properties, yields the equation
dT (x + ∆x, t)
A∆xρCp =
dt

∂T (x, t) ∂T (x + ∆x, t)
= v(t)AρCp [T (x, t) − T (x + ∆x, t)] + Ak − + (1.16)
∂x ∂x
The following issues explain the equation:
• We have assumed the volume A∆x to behave as if it were ideally mixed, i.e., its
bulk temperature equals the local temperature T (x + ∆x, t) occurring at the outflow
location
• In the right hand side we have two terms. The first one describes convective transport
due to particle velocity v(t)
• The second term describes diffusion. The occurrence of heat conduction or molecular
diffusion introduces a heat flux assumed to be proportional to the spatial tempera-
ture gradient. The representation of this phenomenon can also be used to represent
turbulent eddy diffusion.
For the diffusion phenomena, we assume a transport term across the boundary to be pro-
portional to the negative of the local spatial temperature gradient:
∂T (x, t)
− (1.17)
∂x
with a proportionality factor given by the thermal conductivity k (kgm/s3 deg). Division
of the equation (1.16) by A∆xρCp , and letting ∆x → 0, leads to
∂T (x, t) ∂T (x, t) ∂ 2 T (x, t)

+ v(t) = DT (1.18)
∂t ∂x ∂x2
where the thermal diffusion coefficient DT is defined by equation (1.14). The derived one-
dimensional plug-flow based energy balance equation (1.18) combines the phenomena of
convection and diffusion and is called a convection-diffusion equation. This can be shown
by first investigating the case DT = 0. In that case we have the one-dimensional microscopic
balance for convective transport
∂T (x, t) ∂T (x, t)
+ v(t) =0 (1.19)
∂t ∂x
with initial- and boundary conditions
T (x, t0 ) = Tinit (x) (1.20)

T (x0 , t) = Ti (t) (1.21)
Equation (1.20) specifies the initial temperature profile to be given for all x ≥ x0 in the
spatial domain, e.g., x ∈ [0, L] if x0 = 0 is at the inlet and x = L is at the outlet position
of the considered pipe flow system. Equation (1.21) specifies the inlet temperature as a
function of time for all t ≥ t0 .
A different simplification occurs if we assume the velocity to be zero, v(t) = 0, in
equation (1.18). In that case we obtain the one-dimensional diffusion equation
∂T (x, t) ∂ 2 T (x, t)
= DT (1.22)
∂t ∂x2
for which we need to specify initial- and boundary conditions, which now will involve
T (x, t), ∂T∂x
(x,t)
or both. The issues of how to specify initial- and boundary conditions for
these partial differential equations is important. In a future chapter, the notion of bilaterally
coupled systems will be considered. With respect to boundary conditions for the diffusion
equation this implies that at each side of the spatial domain (i.e., at x = 0 and at x = L),
one can specify either a temperature or a heat flux as boundary condition (or a single linear
combination of these two), but not both. Specifying a heat flux at a boundary is equivalent
to specifying ∂T∂x (x,t)
at that boundary.
The terminology for the models defined in this section is as follows. Microscopic balances
are written in terms of partial differential equations. A system described by a partial
differential equation is called a distributed parameter system or an infinite-dimensional
system. A system described by a finite number of integral balances, i.e., macroscopic
balances, is called a lumped parameter system, and its mathematical representation is in
terms of ordinary differential equations.
1.5. LINEARITY AND LINEARIZATION OF NONLINEAR MODELS 19
1.5 Linearity and Linearization of Nonlinear Models

Suppose that a lumped parameter system is described by the possibly nonlinear vector-
differential equation in state-space form
d
x(t) = f (x(t), u(t)) x(t0 ) = x0 (1.23)
dt
where x, f ∈ Rn , u ∈ Rx m. The m variables in the vector u(t) are the input variables, to
which the system is assumed to be subjected by the surroundings. The n variables in x(t)
are called the state variables. The n model relations expressed by f (x, u) are assumed to
be constant or time-invariant.
It can be observed that most processes behave globally as nonlinear systems, whereas many
processes behave locally as approximately linear systems. This observation provides the
motivation for deriving linearized models.
1.5.1 Definition of linearity

Linearity is a property which is identical to the fact that the superposition principle applies.
The model (1.23) is linear if it satisfies the following three basic superposition properties:
1. Zero-state linearity:
x(φx , t0 , au1 + bu2 ) ≡ ax(φx , t0 , u1 ) + bx(φx , t0 , u2 ) (1.24)
for all inputs u1 ,u2 ∈ Rm and for all constants a, b ∈ R, where φx denotes the zero
state of the system.
2. Zero-input linearity:
x(ax01 + bx02 , t0 , φu ) ≡ ax(x01 , t0 , φu ) + bx(x02 , t0 , φu ) (1.25)
for all states x01 , x02 ∈ Rn and constants a, b ∈ R, where φu denotes the zero input
function.
3. Decomposition property:
x(x0 , t0 , u) = x(φx , t0 , u) + x(x0 , t0 , φu ) (1.26)
for all x0 ∈ Rn and all inputs u ∈ Rm .
The three conditions in the definition can be given an interpretation as follows. Zero-state
linearity requires superposition to hold for the system when the initial state is zero and
a state response occurs only due to the effect of an input. Similarly, zero-input linearity
requires the system to obey superposition in the response to non-zero initial states, when
the inputs are zero. Finally, the decomposition property requires a state response to be the
sum of the effects of a response due to an input for initial state zero, and a response due
to a non-zero initial state for inputs equal to zero.
Now a nonlinear system is a system for which we only specify that the superposition
principle does not hold. This is a rather void characterization in the sense that it states
the fact that an idealised property such as linearity does not hold, without stating any
further characterization or property of the system. Thus we may expect that a further
characterization of the properties of a nonlinear system is required before we can make any
assessment about its behaviour.
We may think of linear systems as locally linearised approximations of smooth nonlinear
processes. There are some strong arguments in favour of the study of linear systems :
• For linear systems, analytic solutions and analytically derived system properties are
available. It is often useful to determine analytically the parameter groups that
are responsible for such properties as steady-state gain, time constants, coupling
coefficients, stability behaviour, or sensitivity coefficients with respect to parameter
variations.
• For nonlinear systems, such an analysis is in general not possible. We have to resort
to numerical simulations and numerical computations for nonlinear systems. A com-
puter can provide a numerical solution, but it in general does not provide background
information about the relationship of this solution to the parameters of the model.
• For many processes the function f (x, u) in the right hand side of the model (1.23)
is a smooth function of x and u, except possibly in a finite number of values of its
arguments. This implies that local linearization might be feasible in many cases.
• Many of the presently available techniques for stability analysis and for feedback
design for nonlinear systems are direct extensions of corresponding linear techniques
and they require smoothness of f (x, u) in equation (1.23). Thus a linearized analysis
provides a result that also is relevant to the understanding of the local behaviour of
the nonlinear techniques.
On the other hand, we must keep in mind proper respect for the limitations of a lin-
earized analysis. For example, there exist certain stability issues for the global behaviour of
nonlinear systems for which a linearized analysis does not present any valuable information.
1.5.2 Linearization
Let x∗ , u∗ be the state and input corresponding to a set of stationary operating conditions
for the nonlinear state-space model of equation (1.23). A necessary and sufficient condition
for x∗ , u∗ to qualify in defining stationary operating conditions is that they satisfy
0 = f (x∗ , u∗ ) (1.27)
In general there are infinitely many x∗ , u∗ satisfying equation (1.27). Now consider
x(t) = x∗ + ∆x(t) u(t) = u∗ + ∆u(t) (1.28)
and expand the function f (x, u) in a Taylor series about the nominal values x∗ , u∗ :

∗ ∗ ∂f ∂f
f (x, u) = f (x , u ) + ∆x + ∆u + higher-order terms (1.29)
∂x ∗ ∂u ∗
Because f (x∗ , u∗ ) satisfies (1.27), we have

d ∂f ∂f
∆x(t) = ∆x(t) + ∆u(t) + higher-order terms (1.30)
dt ∂x ∗ ∂u ∗
For sufficiently smooth f and small ∆x(t), ∆u(t), the higher-order terms are small compared
to the linear terms and can be neglected. By doing this we approximate the nonlinear
equations by linear ones and this leads to the approximating linear state-space model
d
∆x(t) = A∆x(t) + B∆u(t) ∆x(t0 ) = ∆x0 (1.31)
dt
where

∂fi ∂fi
Aij = , Bik = (1.32)
∂xj ∗ ∂uk ∗
i, j ∈ {1, 2, · · · , n} k ∈ {1, 2, . . . , m}
define the matrices

∂f ∂f
A= ∈ Rn×n , B= ∈ Rn×m (1.33)
∂x ∗ ∂u ∗
If our nominal conditions f (x∗ , u∗ ) are not stationary, but rather constitute a nominal tra-
jectory for the system, the same concept of linearization still applies, be it now linearization
along a (given) nominal trajectory. A set of values x∗ (t), u∗ (t) is a nominal trajectory if
x∗ (t), u∗ (t) satisfy the model (1.23). Then the formal execution of the same steps leads to
the linearized model
d
∆x(t) = A(t)∆x(t) + B(t)∆u(t) ∆x(t0 ) = ∆x0 (1.34)
dt
Observe that the linearized model now has time-varying coefficients. In the derivation of
this result we define x(t) = x∗ (t) + ∆x(t), u(t) = u∗ (t) + ∆u(t) so that now ∆x(t), ∆u(t)
denote small perturbations of the actual trajectory of the system about the nominal tra-
jectory x∗ (t), u∗ (t). As an example, such a nominal trajectory can be the trajectory that
is foreseen in a startup or shutdown operation, or during the transfer of a process from
a set of existing operationg conditions to a different one, e.g., grade change. We assume
that a nominal trajectory x∗ (t), u∗ (t) is available in quantitative form. Due to its inherent
variation in time, the partial derivatives

∂f ∂f
, (1.35)
∂x ∗ ∂u ∗
no longer are constant but vary when passing along the trajectory. Consequently, these
matrices can be expressed as a function of time and lead to time-varying matrices A(t), B(t)
in equation (1.33). Note that a time-varying system is the result of a linearization about
a non-constant nominal trajectory. If the nominal trajectory is not varying in time, then
the resulting matrices A, B (1.33) are time-invariant.
1.5.3 Example: Process dynamics of mixing process

Consider the mixing process shown in figure 1.16 where two flows φ1 (t) and φ2 (t) having
concentration c1 (t) and c2 (t), respectively, are input flows for a vessel and are mixed. These
concentrations describe a single component occurring in both flows. The incoming flows
are controlled by a valve in each of the flow lines. Outflow occurs under the influence
of gravity through an orifice. The vessel is agitated by a stirrer. The basic approach of
modelling leads to the following steps.
1. Defining the problem. The purpose of the model is to describe how the outgoing
flow φ(t) and its concentration c(t) depend upon the flows φ1 (t) and φ2 (t) and their
concentrations c1 (t) and c2 (t).
2. System boundary. The mixing process takes place in the interior of the vessel;
take system boundary around the vessel.
1 2
φ1 (t) φ2 (t)
C1 (t) C2 (t)
A
h(t)
S
C(t)
φ(t) C(t)
Figure 1.16: Mixing vessel having natural outflow through orifice
inlet flow phi2 outlet concentration

1.5
0.5
0.45
1
kmol/m3
m3/hr
0.4
0.5
0.35
0.3
0
0 10 20 30 0 10 20 30
outlet flow phi level

2.5
1.6
1.4 2
1.2
1.5
m3/hr
1
m
0.8 1
0.6
0.5
0.4
0.2 0
0 10 20 30 0 10 20 30
time (hrs) −−> time (hrs) −−>
Figure 1.17: Responses of mixing process on inlet flow step

inlet flow phi2 outlet concentration
1.26 0.3006
b
a
0.3004
1.255
0.3002
kmol/m3
m3/hr
1.25 0.3
0.2998
1.245
0.2996
b
1.24 a
0.2994
0.2992
0 5 10 15 0 5 10 15
outlet flow phi level
1.51 2.28
a a
1.505
2.26
m3/hr
m
1.5
2.24
1.495
b b
1.49 2.22
0 5 10 15 0 5 10 15
time (hrs) −−> time (hrs) −−>
Figure 1.18: Small signal responses of mixing process on inlet flow step
3. Variables through system boundary. The flows φ1 (t) and φ2 (t) are the result
of pressure drop and valve openings in the surroundings, and thus are inputs to our
system. Their concentrations c1 (t) and c2 (t) describe properties of mass flow particles
that pass the system boundary inwards, and thus also have to be considered as input
variables. By similar reasoning, φ(t) and c(t) are to be considered as output variables.
4. Model hypotheses.
• The vessel is assumed to be ideally mixed. The concentration in the vessel is
not space-dependent and thus can be represented by a single variable c(t).
• We assume that no leakage or evaporation of the fluid occurs.
• The density ρ is assumed to be independent of the concentration c(t).
• The cross-sectional area A is constant.
• The level in the vessel can be described by a single variable h(t).
5. Development of a conceptual model. The total volumetric incoming and outgo-
ing flows will determine the level. The outgoing flow will be related to the pressure
drop over the orifice which has a fixed opening. This pressure drop is proportional to
the level in the vessel.
6. Decomposition into subsystems. As the valves are left outside the system bound-
ary, there is no direct reason to partition the system into subsystems.
7. Formulation of conservation laws. In the problem we have the 7 variables φ1 (t),
φ2 (t), c1 (t), c2 (t), φ(t), c(t) and h(t). Of these 7 variables, 4 are given by the sur-
roundings as inputs. Thus we need 3 independent equations to describe the remaining
3 variables. We have the following relations:
mass balance (total):
dρAh(t)
= ρφ1 (t) + ρφ2 (t) − ρφ(t) (1.36)
dt
mass balance (component):
dc(t)Ah(t)
= c1 (t)φ1 (t) + c2 (t)φ2 (t) − c(t)φ(t) (1.37)
dt
flow through orifice:

φ(t) = k h(t) (1.38)
where k is a constant, determined by the opening of the orifice and by material

properties of the fluid. The three model relations (1.36), (1.37) and (1.38) describe
the model. There are two differential equations and one algebraic equation.
8. Mathematical simplifications. The three relations (1.36), (1.37) and (1.38) can
be simplified by eliminating φ(t) from (1.37) by using (1.38), and by inserting (1.36)
into (1.37):
dh(t)
A = φ1 (t) + φ2 (t) − k h(t) (1.39)
dt
dc(t)
Ah(t) = [c1 (t) − c(t)]φ1 (t) + [c2 (t) − c(t)]φ2 (t) (1.40)
dt
We now have two model equations in the form of two differential equations, with
h(t) and c(t) as unknowns. The equation for the output flow (1.38) can be part of
the model in the sense that it does not describe the internal variables of the system,
but, given the internal variables, it can be considered as defining an output variable,
i.e., a variable that is transmitted through the system boundary to the surroundings.
Simulation results in figures 1.17 and 1.18 show typical responses of the process. The
following parameter values have been used:
A = 1.5 m2 k = 1.0
c∗1 = 0.8 kmol/m3 φ∗1 = 0.25 m3 /hr
c∗2 = 0.2 kmol/m3 φ∗2 = 1.25 m3 /hr
The figure 1.17 shows process responses for a large step disturbance in inlet flow
φ2 . Note the difference in concentration response speed for the high level and low
level situation. The responses show nonlinear behaviour, as superposition clearly
does not hold. The figure 1.18 shows small signal behaviour in the close vicinity of a
stationary operating condition. The process variables respond almost as if the process
were linear. The stationary initial process condition in both figures is identical.
9. Mathematical approximation. If desired, the model relations 1.40 and 1.39 can
be linearized about a set of stationary operating conditions. Write each variable as
h(t) = h∗ + ∆h(t)
c(t) = c∗ + ∆c(t)
and similarly for other variables. Here, the variables h∗ , c∗ indicate the value of the
variable in the stationary operating condition, and ∆h(t), ∆c(t) are the deviations
from these values. Inserting these variables into equation (1.40) yields:
d[c∗ + ∆c(t)]
A[h∗ + ∆h(t)] = [c∗1 + ∆c1 (t) − c∗ − ∆c(t)][φ∗1 + ∆φ1 (t)]+
dt
[c∗2 + ∆c2 (t) − c∗ − ∆c(t)][φ∗2 + ∆φ2 (t)] (1.41)

In steady-state the relation is:
0 = [c∗1 − c∗ ]φ∗1 + [c∗2 − c∗ ]φ∗2 (1.42)
Subtracting (1.42) from (1.41):
d∆c(t) d∆c(t)
Ah∗ + A∆h(t) = [c∗1 − c∗ ]∆φ1 (t) + φ∗1 [∆c1 (t) − ∆c(t)]
dt dt
+[∆c1 (t) − ∆c(t)]∆φ1 (t) + [c∗2 − c∗ ]∆φ2 (t) + φ∗2 [∆c2 (t) − ∆c(t)]
+[∆c2 (t) − ∆c(t)]∆φ2 (t) (1.43)
Linearization of (1.43) amounts to deleting all terms that contain a product of two
∆-variables:
d∆c(t)
Ah∗ = [c∗1 − c∗ ]∆φ1 (t) + φ∗1 [∆c1 (t) − ∆c(t)]+
dt
[c∗2 − c∗ ]∆φ2 (t) + φ∗2 [∆c2 (t) − ∆c(t)] (1.44)
A similar operation can be performed by inserting the perturbed variables into the
equation (1.39), yielding:
d[h∗ + ∆h(t)]
A = φ∗1 + ∆φ1 (t) + φ∗2 + ∆φ2 (t) − k h∗ + ∆h(t) (1.45)
dt
In the steady state the relation is:
√
0 = φ∗1 + φ∗2 − k h∗ (1.46)
The term containing the square root can be linearized by the following observation:

√ ∆h(t)
h∗ + ∆h(t) = h∗ 1+
h∗
√ ∆h(t)
= h∗ [1 + ] + higher order terms in ∆h(t) (1.47)
2h∗
The last step can be verified by computing the square of the left and right hand sides
of equation (1.47). Inserting (1.46) and (1.47) into (1.45) yields the second linearized
model relation:
d∆h(t) φ∗ ∆h(t)
A = ∆φ1 (t) + ∆φ2 (t) − (1.48)
dt 2h∗
where φ∗ = φ∗1 + φ∗2 . The linearized version of (1.38) is:
√ ∆h(t)
φ∗ + ∆φ(t) = k h∗ [1 + ] (1.49)
2h∗
so that
φ∗
∆φ(t) = ∆h(t) (1.50)
2h∗
Define the parameter θ as the average residence time of mass particles in the vessel:
h∗ A
θ= (1.51)
φ∗
Then the equations (1.44), (1.48) and (1.50) can be formulated in the standard state
space form for linear sets of differential equations:
 
∆φ1 (t)
1 1 1  ∆φ2 (t) 
d ∆h(t) − 2θ 0 ∆h(t) A A 0 0  
= + c∗1 −c∗ c∗2 −c∗ φ∗1 φ∗2  ∆c1 (t) 
dt ∆c(t) 0 − 1θ ∆c(t) Ah∗ Ah∗ Ah∗ Ah∗
∆c2 (t)
A

∆φ(t) 2θ 0 ∆h(t)
= (1.52)
∆c(t) 0 1 ∆c(t)
The model is in the standard form

d
x(t) = Ax(t) + Bu(t)
dt
y(t) = Cx(t) (1.53)
where x(t) is the state vector, u(t) is the input vector, and y(t) is the output vector.
1.6 Differential-algebraic model equations

The model in equation (1.23) is not the most general one encountered in dynamic modelling.
In fact, in general we have coupled differential equations and algebraic equations of the form
d
x1 (t) = f (x1 (t), x2 (t), u(t))
dt
0 = g(x1 (t), x2 (t), u(t)) (1.54)
The algebraic equations (1.54) are in implicit form. This may be experienced by the
occurrence of algebraic loops, i.e., implicit algebraic relations amongst the variables in
x1 , x2 and u. We can try to replace the equations (1.54) by an equivalent explicit set of
equations in which the variables x2 (t) are specified as explicit functions of x1 (t) and u(t) :
x2 (t) = h(x1 (t), u(t)) (1.55)
If indeed the model (1.54) can be brought into the format (1.55), we can insert (1.55) into
the differential equations in (1.54) and obtain a model in state-space form:
d
x1 (t) = f (x1 (t), h(x1 (t), u(t)), u(t)) (1.56)
dt
or
d
x1 (t) = f ∗ (x1 (t), u(t)) (1.57)
dt
The structure of these relations is shown in figure 1.19. Note that all signals pass through
the integrator in their route from inputs to outputs.
For smooth functions f, g in equations (1.54), the solutions to the linearized model are
contained in the set of solutions to the nonlinear model, be it in an approximate sense
1.6. DIFFERENTIAL-ALGEBRAIC MODEL EQUATIONS 27
f∗ x1 (0)
u(t)
u(t) dx1 (t)
x2 (t) x1 (t)
h f dt
x1 (t)
Figure 1.19: Differential-algebraic equations, transformed to state-space form
and under the conditions of the validity of these approximations. For the approach shown
above for bringing a model formulated in terms of algebraic and differential equations
into an explicit state-space form, further insight into the existence question of this model
transformation can be derived from a linearized analysis.
A linearized version of f and g with respect to x1 , x2 and u about a stationary solution
x1 , x∗2 , u∗ leads to the linearized model
∗
d
∆x1 (t) = A11 ∆x1 (t) + A12 ∆x2 (t) + B1 ∆u(t)
dt
0 = A21 ∆x1 (t) + A22 ∆x2 (t) + B2 ∆u(t) (1.58)
where

∂f ∂f ∂f
A11 = ∂x1 ∗ A12 = ∂x2 ∗ B1 = ∂u ∗
(1.59)
∂g ∂g ∂g
A21 = ∂x1 ∗ A22 = ∂x2 ∗ B2 = ∂u ∗
Now a sufficient condition for the existence of an explicit expression that is equivalent to
the algebraic equations (1.58) is that the matrix A22 is non-singular, i.e., it is invertible.
In that case, the explicit expression is
∆x2 (t) = −A−1 −1

22 A21 ∆x1 (t) − A22 B2 ∆u(t) (1.60)
Inserting this explicit expression for ∆x2 (t) into the differential equations in (1.58) leads
to the linear state-space model
d
∆x1 (t) = A∆x1 (t) + B∆u(t) ∆x1 (t0 ) = x10 (1.61)
dt
with
A = A11 − A12 A−1

22 A21 (1.62)
B = B1 − A12 A−1
22 B2 (1.63)
If A22 turns out to be singular, then a further analysis is necessary in order to assess the
properties of the system under consideration. In that case we may have to investigate the
index of the system, or it may turn out that our system model is internally or externally
nonproper. The discussion of these issues will be given in chapter 5.
Tset (t)
feed flow
control
TC
Ci (t) Ti (t)
cooling water out
Tj (t) C(t)
X(t)
φj (t) T (t)
jacket
Tjo (t)
cooling water
vessel
product flow
C(t) T (t)
Figure 1.20: Exothermal chemical reactor
The perturbation variables ∆x,∆u will often be replaced by x, u if the context of a

linear model is clear. Then equation (1.61) is written as
d
x1 (t) = Ax1 (t) + Bu(t) x1 (t0 ) = x10 (1.64)
dt
Assuming that A, B are time-invariant, the solution is:
t
A(t−t0 )
x1 (t) = e x10 + eA(t−τ ) Bu(τ )dτ (1.65)
t0
where the matrix exponential is definied by its series expansion

1 2 2
eAt = I + At + A t + ... (1.66)
2!
1.7 Continuous chemical reactor process dynamics

As an example of a more involved process modeling exercise, a continuous flow system
will be considered in which a nonisothermal chemical reaction takes place. The reaction is
assumed to be exothermic and irreversible. Reactant A forms the product B at a specific
k
reaction rate k, which is indicated by A −→ B. The reaction vessel will be assumed to be
well mixed, and such a process is known under the name CSTR (Continuous Stirred-Tank
Reactor). There is a continuous input feed flow to the process, and a continuous outflow.
Density variations are assumed to be neglected, and the vessel is completely filled. Thus
input volumetric flow φ(t) is equal to the output volumetric flow. The flow can vary in
time. The vessel is surrounded by a cooling jacket to remove the heat of reaction. The
amount of cooling water is controlled by a valve, the opening of which is determined by
1.7. CONTINUOUS CHEMICAL REACTOR PROCESS DYNAMICS 29
Tjo (t) Ti (t) Ci (t) φ(t)
Tj (t) ? ? ? C(t)
Tset (t) - - -
- jacket reactor
controller - valve - -
- X(t) φj (t) Q(t) T (t)
Figure 1.21: Block diagram of CSTR subsystems and interconnections
a PI-controller. The reaction temperature in the vessel is the controlled variable. The
process is shown in figure 1.20. The various decisions underlying the modelling procedure
closely follow the scheme given in table 1.1:
1. Intended purpose of the model: To describe the dynamics of the process such
that conclusions regarding stability of operation can be drawn
2. System boundary: The exothermic reaction needs cooling, and the amount of cool-
ing must be adapted to the actual heat production if constant operational conditions
are required in the face of disturbances. Thus the system consists of vessel, cooling
jacket, cooling flow valve and temperature controller. The amount of flow through
the vessel is assumed to be determined outside the system.
3. Variables across system boundary: The following input variables must be con-
sidered:
• volumetric flow φ(t) (m3 /s) through the vessel

• temperature Ti (t) (K) of the input flow to the vessel
• concentration Ci (t) (mol/m3 ) of component A in the input flow
• coolant flow inlet temperature Tjo (t) (K)
• setpoint value Tset (K) for the temperature controller
As internal variables we have:
• volumetric flow φj (t) (m3 /s) through the jacket

• temperature Tj (t) (K) of the jacket
• concentration C(t) (mol/m3 ) of component A in the vessel
• temperature T (t) (K) of the fluid in the vessel
This list must be completed on the basis of a further study of the model hypotheses
and conservation laws.
4. Model hypotheses:
• The vessel is perfectly mixed

• There are no heat losses to the environment from coolant flow, jacket and vessel
• The jacket is assumed to be perfectly mixed
• The heat capacity of the vessel wall between vessel and jacket is neglected; its
heat resistance is assumed to be incorporated in the heat transfer coefficients
between coolant and wall, and between wall and fluid in the vessel
• The heat transfer between coolant and wall, and between wall and fluid in the
vessel is assumed to be proportional to the difference in temperatures
• The vessel fluid volume V is constant
• The density of the fluid ρ (kg/m3 ) is independent of temperature T (t) and
concentration C(t), so that input volumetric flow φi (t) (m3 /s) instantaneously
equals output volumetric flow φ(t) (m3 /s). The input flow is assumed to be
determined outside the system.
• All material properties of the fluid in the vessel are constant, independent of
concentration C(t)
• The pressure drop over the coolant valve is assumed to be constant
5. Subdivision into subsystems: The four components of the process determine four
subsystems which will be discussed regarding variables and interconnection structure.
The variables and interconnection structure are shown in the block diagram of figure
1.21.
(a) Reactor vessel

• input variables: Ci (t), Ti (t), φ(t), Tj (t)
• internal variables: C(t), T (t), heat flow Q(t) (J/s) between vessel and jacket,
rate of reaction R(t) (mol/m3 s)
Thus 4 equations required. The variables C(t), T (t) and Q(t) act as output
variables.
(b) Jacket
• input variables: Tjo (t), φj (t), Q(t)
• internal variable: Tj (t)
Thus 1 equation required. The variable Tj (t) acts as output variable.
(c) Temperature controller
• input variables: Tset (t), T (t)
• internal variables: Temperature error (t) (K), valve position X(t) (m)
Thus 2 equations required. The variable X(t) acts as output variable.
(d) Coolant valve
• input variable: X(t)
• internal variable: φj (t)
Thus 1 equation required. The variable φj (t) acts as output variable.
Note that the jacket and reactor are thermally bilaterally coupled.
6. Conservation laws and other model relations: These will be discussed for each
subsystem separately.
(a) Reactor vessel

Mass balance for the component A:
d
{V C(t)} = φ(t)[Ci (t) − C(t)] − V R(t) (1.67)
dt
Energy balance for the vessel
d
{V ρCp T (t)} = φ(t)ρCp [Ti (t) − T (t)] − Q(t) + (−∆i)V R(t) (1.68)
dt
Here −∆i (J/mol) is the heat of reaction, Cp (J/kgK) is the specific heat. The
equation for the heat flow is:
Q(t) = αAc [T (t) − Tj (t)] (1.69)
where α (J/m2 Ks) is the total heat transfer coefficient between vessel and
coolant in the jacket, and Ac (m2 ) is the total area of heat transfer between
vessel fluid and jacket.
The reaction rate for an nth -order reaction can be described by
E
R = C n ke− RT (1.70)
where k is the frequency factor, E (J/kg) is the activation energy of the reaction,
and R (J/kgK) is the specific gas constant. In our case where one species A
reacts with a component assumed to be available in excess, the reaction is of
first order, i.e., n = 1.
(b) Jacket
Energy balance for the jacket volume:
d
{Vj ρj Cpj Tj (t)} = φj (t)ρj Cpj [Tjo (t) − Tj (t)] + Q(t) (1.71)
dt
where Vj (m3 ) is the jacket volume, Cpj (J/kgK) is the specific heat of the
coolant, ρj (kg/m3 ) is the coolant density.
(c) Temperature controller
Suppose that this controller has the charateristics of a PI-controller:
(t) = Tset (t) − T (t) (1.72)
t
X(t) = Kp (t) + Ki (τ )dτ (1.73)
−∞
Here, Kp and Ki are the controller parameters for proportional action and for
integral action, respectively.
(d) Coolant valve
If we assume a linear relationship between valve position and flow area, we have:


 = Kv X(t) if 0 ≤ X ≤ Xmax

φj (t) = 0 if X < 0 (1.74)


 = K X if X > X
v max max
where Xmax is the value for X(t) for which the valve is maximally open, and Kv
is a normalization constant allowing to take Xmax = 1.
-F
G, -F
T∗
Figure 1.22: Heat generation G(T ∗ ) and removal −F(T ∗ ) in CSTR
7. Model structure investigation First the structure of the set of differential and
algebraic equations comprising the model will be analysed. The open-loop model of
the reactor and jacket (without controller and coolant valve) will be considered. The
dynamic balance equations contain time derivative of the accumulation terms in the
left-hand side. These terms will be brought to a form containing only time derivatives
of individual variables. These variables will turn out to become the state variables in
the nonlinear dynamic system.
d φ(t)
C(t) = [Ci (t) − C(t)] − R(t) (1.75)
dt V
d φ(t) Q(t) (−∆i)
T (t) = [Ti (t) − T (t)] − + R(t) (1.76)
dt V V ρCp ρCp
d φj (t) Q(t)
Tj (t) = [Tjo (t) − Tj (t)] + (1.77)
dt Vj Vj ρj Cpj
0 = Q(t) − αAc [T (t) − Tj (t)] (1.78)
− RTE(t)
0 = R(t) − C(t)ke (1.79)
The equations (1.75) - (1.79) are in the form of the general set of differential-algebraic
equations discussed earlier in section 1.5, equations (1.54):
d
x1 (t) = f (x1 (t), x2 (t), u(t))
dt
0 = g(x1 (t), x2 (t), u(t)) (1.80)
In the present case, the vectors x1 (t), x2 (t) and u(t) are:
 
  Tjo (t)
C(t)  φj (t) 
Q(t)  
x1 (t) =  T (t)  x2 (t) = u(t) = 
 Ti (t) 
 (1.81)
R(t)  
Tj (t) Ci (t)
φ(t)
The set of equations (1.78, 1.79) determining x2 (t) can be written explicitly as
Q(t) = αAc [T (t) − Tj (t)]

− RTE(t)
R(t) = C(t)ke (1.82)
and inserted in the equations (1.75) - (1.77), leading to the state space equations:
d φ(t) − E
C(t) = [Ci (t) − C(t)] − C(t)ke RT (t) (1.83)
dt V
d φ(t) αAc [T (t) − Tj (t)] (−∆i) − E
T (t) = [Ti (t) − T (t)] − + C(t)ke RT (t)(1.84)
dt V V ρCp ρCp
d φj (t) αAc [T (t) − Tj (t)]
Tj (t) = [Tjo (t) − Tj (t)] + (1.85)
dt Vj Vj ρj Cpj
The elimination of two internal variables using the two algebraic equations of the
model exactly corresponds to the elimination procedure described in section 1.5,
leading to equations (1.56, 1.57).
8. Steady-state nonlinearity and stability analysis The state-space model (1.83

- 1.85) is nonlinear for several reasons. One is the occurrence of products of input
and state variables in the right-hand sides of the state space equations, e.g., the
product φ(t)C(t) in equation 1.83. Another reason is the appearance of the nonlinear
expression for the reaction rate. The character of this nonlinearity will be investigated
by considering the steady-state behaviour of the reactor vessel. The reactor vessel
has as inputs the variables Ci (t), Ti (t), φ(t) and Tj (t). In steady state, the equations
(1.83), (1.84) are
E
φ∗ [Ci∗ − C ∗ ] − V C ∗ ke− RT ∗ = 0 (1.86)
E
φ∗ ρCp [Ti∗ − T ∗ ] − αAc [T ∗ − Tj∗ ] + V (−∆i)C ∗ ke− RT ∗ = 0 (1.87)
where the variables (·)∗ indicate steady-state values. From (1.86) follows an explicit
expression for C ∗ :
φ∗ Ci∗
C∗ = E
(1.88)
φ∗ + V ke− RT ∗
The left-hand side of the energy balance (1.87) can be split into two parts; one part
describes the heat generation G(T ∗ ), the other part is the heat removal from the
vessel −F(T ∗ ):
E
G(T ∗ ) := AH(−∆i)C ∗ ke− RT ∗ (1.89)
Inserting (1.88):
E
∗ φ∗ Ci∗ (−∆i)ke− RT ∗
G(T ) := E (1.90)
φ∗
AH + ke− RT ∗
For given input variables φ∗ , Ci∗ and given material properties and reactor volume
this indeed is only a function of T ∗ . The function G(T ∗ ) is S-shaped and is shown
in figure 1.22 as a function of T ∗ . The other terms in equation (1.87) define the heat
removal −F(T ∗ ):
−F(T ∗ ) = φ∗ ρCp [T ∗ − Ti∗ ] + αAc [T ∗ − Tj∗ ] (1.91)
For given input variables φ∗ , Ti∗ , Tj∗ this is a linear function of T ∗ :
−F(T ∗ ) = [φ∗ ρCp + αAc ]T ∗ − φ∗ ρCp Ti∗ − αAc Tj∗ (1.92)
The function −F(T ∗ ) is also shown in figure 1.22 as a function of T ∗ . In steady

state the heat generation equals the heat removal. The figure shows three possible
steady state equilibrium solutions: one for low temperature, one for intermediate
temperature and one for high temperature. Suppose the process is in the intermediate
point of intersection. Let a small increase in heat generation lead to a slightly higher
temperature, i.e., to a temperature to the right of the intersection point. Then the
heat removal is smaller than the heat generation, i.e., the process tends to be removed
away from the intersection conditions. The steady state operational condition in the
intermediate intersection point is called statically unstable, the other intersections are
statically stable. The following remarks apply.
• A dynamic analysis shows that the process operated in the intermediate inter-
section point is also asymptotically unstable from a dynamic point of view, i.e.,
regarding its natural behaviour under disturbances as a function of time.
• The stability result holds for a fixed jacket temperature Tj∗ . If the jacket tem-
perature is not fixed, but varies as a function of T ∗ , a different heat removal
function results, exhibiting different equilibrium and stability properties.
• In particular if stabilizing proportional temperature feedback control is applied,
then the actual steady-state heat removal function −F(T ∗ ) can be given a
greater slope. If stabilizing integral feedback control of the temperature T ∗
is applied, then the slope of the heat removal function −F(T ∗ ) will be infinite,
so that only one equilibrium intersection point occurs.
9. Dynamic simulation of the CSTR A dynamic simulation of the complete process

shows the behaviour of the process under disturbances of the various inputs, for
different initial steady-state operating conditions. The following numerical values
have been used.
φ = 1.13 m3 /hr k = 7.08 × 1010 hr −1

AH = 1.36 m3 E = 6.96 × 107 J/kmol
Ci = 8.0 kmol/m3 R = 8314.3 J/kmolK
Ti = 294 K α = 3.07 × 106 J/m2 hrK
Tjo = 294 K Ac = 23.2 m2
(−∆i) = 6.96 × 107 J/kmol Vj = 0.109 m3
ρ = 801 kg/m3 ρj = 998 kg/m3
Cp = 3140.1 J/kgK Cj = 4186.8 J/kgK
Kp = −0.5 Xmax = 1
Ki = −5 Kv = 2.5
The results are shown in the figures (1.23 . . . 1.29).

Steady state reactor temperature vs coolant flow

420
400
380
T (degrees)
360
340
320
300
0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6
coolant flow (m3/hr)
Figure 1.23: Steady state process operational curve of CSTR
setpoint Tset concentration C
4.4
335
4.2
kmol/m3
degrees
330 4
3.8
325
3.6
0 2 4 6 8 0 2 4 6 8
temperature T coolant flow
335 2
degrees
m3/hr
330 1
325 0
0 2 4 6 8 0 2 4 6 8
coolant temperature Tj reaction rate script−R
5
335
kmol/m3hr
degrees
4
330
3
325
2
0 2 4 6 8 0 2 4 6 8
Figure 1.24: Process response on step in Tset under closed-loop control

coolant flow coolant flow

0.75 0.65
a a
0.645
0.7
0.64
m3/hr
m3/hr
0.65
0.6 0.635
b 0.63 b
0.55
0 1 2 3 0 1 2 3
temperature T temperature T
330 312
325 b 311.5
degrees
degrees
320 311 b
315 310.5
310 310
a
a
305 309.5
0 1 2 3 0 1 2 3
Figure 1.25: Responses on step in coolant flow, open loop T ∗ = 310 K
coolant flow coolant flow
1.5 a 1.418 a
1.45 1.4175
m3/hr
m3/hr
1.4 1.417
1.35 1.4165
b b
1.3 1.416
0 1 2 3 4 5 0 1 2 3 4 5
temperature T temperature T
360.4
370 b
360
360.2 b
350
degrees
degrees
340
360
330
320
359.8 a
310
a
300
359.6
0 1 2 3 4 5 0 1 2 3 4 5
Figure 1.26: Responses on step in coolant flow, open loop T ∗ = 360 K

coolant flow coolant flow coolant flow

0.6385 0.9365
0.8935
0.638 0.936
0.893
m3/hr
0.6375 0.9355
0.8925
0.637 0.935
0.892
0.6365 0.9345
0 5 10 0 5 10 0 5 10
temperature T temperature T temperature T

300.3 310.3 320.3
300.2 310.2 320.2
300.1 310.1 320.1

degrees
300 310 320
299.9 309.9 319.9
299.8 309.8 319.8

0 5 10 0 5 10 0 5 10
time (hrs) time (hrs) time (hrs)
Figure 1.27: Responses on step in coolant flow, open loop T ∗ = 300, 310, 320 K

1.538
1.296
1.518
1.5375
1.2955
1.5175
m3/hr
1.537
1.295
1.517
1.5365
1.2945
1.5165
1.536
0 5 10 0 5 10 0 5 10

330.3 340.3 350.3
330.2 340.2 350.2
330.1 340.1 350.1

degrees
330 340 350
329.9 339.9 349.9
329.8 339.8 349.8

0 5 10 0 5 10 0 5 10

1.2415
1.0605
1.4175
1.241
1.06
1.417
m3/hr
1.2405
1.0595
1.4165
1.24
1.059
1.416
1.2395
0 5 10 0 5 10 0 5 10

360.3 370.3 380.3
360.2 370.2 380.2
360.1 370.1 380.1

degrees
360 370 380
359.9 369.9 379.9
359.8 369.8 379.8

0 5 10 0 5 10 0 5 10
Ti , Ci , φ
Tj1
Q1
T1 , C1 Q2
Tj1 Tj2
Q3
Tjo T2 , C2 Tj2
Q4
T2 , C2
Figure 1.30: Block diagram of 2 × 2 compartments CSTR model

Steady state reactor temperature vs coolant flow

420
400
380 2 x 2 compartments
T (degrees) 360
perfect mixing
340
320
300
0.6 0.8 1 1.2 1.4 1.6

coolant flow (m3/hr)
Figure 1.31: Steady state process operational curve of compartimental CSTR
setpoint Tset concentration C1
335 5.6
kmol/m3
degrees
5.4
330
5.2
325
5
0 2 4 6 8 0 2 4 6 8
temperature T1 coolant flow
335 2
degrees
m3/hr
330
1
325
0
0 2 4 6 8 0 2 4 6 8
coolant temperature Tj1 reaction rate script−R1
335 6
kmol/m3hr
degrees
330 5
4
325
3
0 2 4 6 8 0 2 4 6 8
Figure 1.32: Compartmental CSTR response on step in Tset , closed-loop control

10. Relaxing the ideally mixing assumptions The simulation model has been ex-
tended by assuming a subdivision into two compartments for both the reaction vessel
and the jacket. Each compartment is assumed to be ideally mixed, and both com-
partments are connected in series without backflow. The heat exchanging area is
subdivided into four equal parts, according to the block scheme in figure 1.30. The
model equations have been formulated for each compartment separately, taking into
account the partial volumes and heat exchange areas. The steady state process op-
erational curve for the compartmental model is compared with the original model in
figure 1.31. Relaxing the ideally mixing assumption has some influence on quanti-
tative steady state and dynamic characteristics, although the qualitative behaviour
is retained. Figure 1.32 shows the responses for the compartmental model under
closed-loop control with step disturbance on temperature setpont. The responses
show considerable agreement in their form with those of figure 1.24, although the
steady state values differ.
1.8 Literature for this Chapter

Basic technical and philosophical ideas of modelling as given in section 1.1 can be found in
Kalman [27], Kwatny and Mablekos [29], [30], Golomb [19], Profos [42] or Bub and Lugner
[10]. A system theory that defines physical models without a priori distinction between
inputs and outputs has been proposed by Willems [52]; see also the earlier work [51], and
the full treatment in Polderman and Willems [39].
The formulation of conservation laws for the general microscopic and macroscopic situations
and for a variety of boundary conditions and coordinate frame descriptions are discussed
in Himmelblau and Bischoff [23] and in Bird et al. [9]. The two volumes (in Dutch) on
transport phenomena by Van den Akker and Mudde [46] and by Hoogendoorn and Van der
Meer [25] provide background on the formulation of basic relations for process models.
The details of the modelling of process heat and mass transfer can be found in books like
Welty, Wicks and Wilson [49] or Hewitt, Shires and Bott [22].
Introductory treatments of process modeling are Guy [20], Roffel and Rijnsdorp [44], and
Kecman [28]. The role of dynamic modelling in control engineering is considered in Profos
[43]. Ljung and Glad [31] consider some elementary dynamic modelling issues in their re-
lationship with system identification.
Modelling from a fundamental chemical engineering point of view is considered in Eigen-
berger [14], Marquardt [34] and Marquardt and Zeitz [36]. Systematic approaches for model
development can be found in Lohmann and Marquardt [32], Marquardt [35], and of phase
equilibrium process modeling are addressed in Ponton and Gawthrop [40], and an approach
towards hybrid models is provided in Barton and Pantelides [8]. A systematic way to for-
mulate process models has been developed in Preisig and Kimmich and Rippin [41], Weiss
and Preisig [48], and Westerweele [50].
The basic techniques of chemical process modelling are treated in Hartmann and Kaplick
[21], Luyben and Wenzel [33], and in Ingham, Dunn and Heinzle [26]. Process modelling
for process control is treated in Ogunnaike and Ray [37]. Classical references on process
modelling are Franks [16], [15], Friedly [17], Aris [1], Aris [2] and Denn [12], [13].
Further discussions can be found in Paynter [38], Asbjornsen [5], [6], Sargent [45], and
Georgakis [18]. The continuous stirred tank reactor has been studied in the classical work
of Aris and Amundson [3], [4]. The multiplicity of steady states has been studied by many
researchers, e.g., Hlavacek, Kubicek and Visnak [24], Varma and Amundson [47], Balako-
taiah and Luss [7], and Cicarelli and Aris [11].
1.8. LITERATURE FOR THIS CHAPTER 41
References
[1] R. Aris. Mathematical Modelling Techniques. Pitman Publishing Ltd., London, UK,
1978.
[2] R. Aris. Method in the modeling of chemical engineering systems. In C. T. Leondes,

editor, Control and Dynamic Systems, volume 15, pages 41–98. Academic Press, New York,
NY, 1979.
[3] R. Aris and N. R. Amundson. An analysis of chemical reactor stability and control. I.
The possibility of local control, with perfect or imperfect control mechanisms. Chemical
Engineering Science, 7:121–131, 1958a.
[4] R. Aris and N. R. Amundson. An analysis of chemical reactor stability and control. II.
The evolution of proportional control. Chemical Engineering Science, 7:132–147, 1958b.
[5] O. A. Asbjornsen. Modeling techniques: theory and practice. Modeling Identification

and Control, 6:105–125, 1985.
[6] O. A. Asbjornsen. A systems engineering approach to process modeling. In D. M. Prett

and M. Morari, editors, The Shell Process Control Workshop. Process Control Research:
Industrial and Academic Perspectives, Houston TX, dec. 15-18, 1986., pages 139–182. But-
terworth Publ., Boston, MA, 1987.
[7] V. Balakotaiah and D. Luss. Analysis of the multiplicity patterns of a CSTR. Chemical
Engineering Communications, 13:111–132, 1981.
[8] P. I. Barton and C. C. Pantelides. Modeling of combined discrete/continuous processes.

AIChE Journal, 40:966–979, 1994.
[9] R. B. Bird, W. E. Stewart, and E. N. Lightfoot. Transport Phenomena. John Wiley

and Sons, Inc., New York, NY, 1960.
[10] W. Bub and P. Lugner. Systematik der Modellbildung. Teil 1: Konzeptionelle Mod-
ellbildung. In I. Troch, editor, Modellbildung für Regelung und Simulation: Methoden-
Werkzeuge-Fallstudien, GMA Aussprachetag, Langen, BRD, 25-26 März 1992. VDI Ber.
Nr. 925., pages 1–18. VDI Verlag, Düsseldorf, 1992.
[11] P. Cicarelli and R. Aris. Continuous reactions in a non-isothermal CSTR- I. Multi-

plicity of steady states. Chemical Engineering Science, 49:621–631, 1994.
[12] M. M. Denn. Modeling for process control. In C. T. Leondes, editor, Control and
Dynamic Systems, volume 15, pages 147–194. Academic Press, New York, NY, 1979.
[13] M. M. Denn. Process Modeling. Longman Scientific and Technical, Harlow, Essex,
UK, 1986.
[14] G. Eigenberger. Modelling and simulation in industrial chemical reaction engineering.

In K. H. Ebert and P. Deuflhard, editors, Modelling of Chemical Reaction Systems. Proc.
Int. Workshop, Heidelberg, FRG, September 1-5, 1980., pages 284–304. Springer Verlag,
Berlin, 1981.
[15] R. G. E. Franks. Mathematical Modeling in Chemical Engineering. John Wiley and

Sons, Inc., New York, NY, 1967.
[16] R. G. E. Franks. Modeling and Simulation in Chemical Engineering. John Wiley and
Sons, Inc., New York, NY, 1972.
[17] J. C. Friedly. Dynamic Behavior of Processes. Prentice Hall, Inc., Englewood Cliffs,
NJ, 1972.
[18] C. Georgakis. On the use of extensive variables in process dynamics and control.
Chemical Engineering Science, 41:1471–1484, 1986.
[19] S. W. Golomb. Mathematical models: uses and limitations. IEEE Transactions on

Reliability, 20:130–131, 1971.
[20] J. L. Guy. New CE Refresher: Process Dynamics. McGraw-Hill Book Co., New York,
NY, 1983.
[21] K. Hartmann and K. Kaplick. Analysis and Synthesis of Chemical Process Systems.
Elsevier Publ., Amsterdam, The Netherlands, 1990.
[22] G. F. Hewitt, G. L. Shires, and T. R. Bott. Process Heat Transfer. CRC Press, Boca
Raton, FL, USA, 1994.
[23] D. M. Himmelblau and K. B. Bischoff. Process Analysis and Simulation. Deterministic

Systems. John Wiley and Sons, Inc., New York, NY, 1968.
[24] V. Hlavacek, M. Kubicek, and K. Visnak. Modelling of chemical reactors XXVI.

Multiplicity and stability analysis of a continuous stirred tank reactor with exothermic
consecutive reactions A → B → C. Chemical Engineering Science, 27:719–742, 1972.
[25] C. J. Hoogendoorn and T. H. Van der Meer. Fysische Transportverschijnselen II.

Delftse Uitgevers Maatschappij, Delft, Nederland, 1991.
[26] J. Ingham, I. J. Dunn, and E. Heinzle. Chemical Engineering Dynamics. Modelling

with PC Simulation. VCH Verlagsgesellschaft, Weinheim, W.Germany, 1994.
[27] R. E. Kalman. Comments on the scientific aspects of modeling. In M. Marois, ed-

itor, Towards a Plan of Actions for Mankind, pages 493–505. North-Holland Publishing
Company, Amsterdam, 1974.
[28] V. Kecman. State-Space Models of Lumped and Distributed Systems, volume 112 of
Lecture Notes in Control and Information Sciences. Springer Verlag, Berlin, 1988.
[29] H. G. Kwatny and V. E. Mablekos. The modeling of dynamical processes. In Proceed-

ings of the 1975 IEEE Conference on Decision and Control, Houston, Texas, USA, 10-12
December 1975., pages 271–281. IEEE, New York, NY, 1975.
[30] H. G. Kwatny and V. E. Mablekos. On the modeling of dynamical processes with

applications in power systems engineering. In L. H. Fink and K. Carlsen, editors, Sys-
tems Engineering for Power: Status and Prospects, Engineering Foundation Conference,
Henniker, NH, USA, August 17-22, 1975., pages 368–392. NTIS, Springfield, 1975.
[31] L. Ljung and T. Glad. Modeling of Dynamic Systems. Prentice Hall, Inc., Englewood
Cliffs, NJ, 1994.
[32] B. Lohmann and W. Marquardt. On the systematization of the process of model

development. Computers and Chemical Engineering, 20:S213–S218, 1996.
1.8. LITERATURE FOR THIS CHAPTER 43
[33] W. L. Luyben and L. A. Wenzel. Chemical Process Analysis: Mass and Energy Bal-
ances. Prentice Hall, Inc., Englewood Cliffs, NJ, 1988.
[34] W. Marquardt. Dynamic process simulation: Recent progress and future challenges.
In Y. Arkun and W. H. Ray, editors, Chemical Process Control - CPC IV, Proceedings of
the 4th Engineering Foundations Conf., South Padre Island, TX, USA, feb. 17-22, 1991.,
pages 131–180. American Institute of Chemical Engineers, New York, 1991.
[35] W. Marquardt. Towards a process modeling methodology. In R. Berber, editor, Meth-

ods of Model Based Process Control, Proceedings of the NATO Advanced Study Institute,
Antalya, Turkey, august 7-17, 1994, pages 3–40. Kluwer Publ., Dordrecht, the Netherlands,
1995.
[36] W. Marquardt and M. Zeitz. Rechnergestützte Modellbildung in der Verfahrenstech-

nik. In I. Troch, editor, Modellbildung für Regelung und Simulation: Methoden-Werkzeuge-
Fallstudien, GMA Aussprachetag, Langen, BRD, 25-26 März 1992. VDI Ber. Nr. 925.,
pages 307–341. VDI Verlag, Düsseldorf, 1992.
[37] B. A. Ogunnaike and W. H. Ray. Process Dynamics, Modeling, and Control. Oxford
University Press, Inc., New York, NY, USA, 1994.
[38] H. M. Paynter. Thermofluid process dynamics in boiler modeling. In D. A. Berkowitz,

editor, Proceedings of the Seminar on Boiler Modeling, Bedford, MA, USA, nov. 6-7, 1974.,
pages 187–206. The MITRE Corporation, Bedford, MA, USA, 1975.
[39] J. W. Polderman and J. C. Willems. Introduction to Mathematical Systems Theory.

A Behavioural Approach. Springer Verlag, New York, NY, 1998.
[40] J. W. Ponton and P. J. Gawthrop. Systematic construction of dynamic models for

phase equilibrium processes. Computers and Chemical Engineering, 15:803–808, 1991.
[41] H. A. Preisig, M. Kimmich, and D. W. T. Rippin. A study of dynamic system mod-

elling. Computers and Chemical Engineering, 12:455–460, 1988.
[42] P. Profos. Some considerations on theoretical process modeling. Trans. ASME /

Journal of Dynamic Systems, Measurement, and Control, 94:282–284, 1972.
[43] P. Profos. Modellbildung und ihre Bedeutung in der Regelungstechnik. In R. Iser-

mann, editor, Prozessmodelle 1977. Modellbildung und Identifikation technischer Prozesse.
Tagung Wiesbaden, 25 und 26 April 1977. VDI-Berichte Nr. 276, pages 5–12. VDI Verlag,
Düsseldorf, 1977.
[44] B. Roffel and J. E. Rijnsdorp. Process Dynamics, Control, and Protection. Ann Arbor
Science Publishers, Ann Arbor, MI, USA, 1982.
[45] R. W. H. Sargent. Advances in modelling and analysis of chemical process systems.

Computers and Chemical Engineering, 7:219–237, 1983/84.
[46] H. E. A. Van den Akker and R. F. Mudde. Fysische Transportverschijnselen I. Delftse

Universitaire Pers, Delft, Nederland, 1996.
[47] A. Varma and N. R. Amundson. Some observations on uniqueness and multiplicity of

steady states in non-adiabatic chemically reacting systems. Canadian Journal of Chemical
Engineering, 51:206–226, 1973.
[48] M. Weiss and H. A. Preisig. Structural analysis in the dynamical modelling of chemical
engineering systems. Mathematical and Computer Modelling of Dynamical Systems, 6:325–
364, 2000.
[49] J. R. Welty, C. E. Wicks, and R. E. Wilson. Fundamentals of Momentum, Heat, and

Mass Transfer. 3rd Edition. John Wiley and Sons, Inc., New York, NY, 1984.
[50] M. R. Westerweele. Five steps for building consistent dynamic process models and
their implementation in the computer tool Modeller. PhD thesis, Technische Universiteit
Eindhoven, 2003.
[51] J. C. Willems. System theoretic models for the analysis of physical systems. Ricerche
di Automatica, 10:71–106, 1979.
[52] J. C. Willems. Paradigms and puzzles in the theory of dynamical systems. IEEE
Transactions on Automatic Control, 36:259–294, 1991.

Basic Modelling of Dynamic Engineering Systems

Uploaded by

Copyright:

Available Formats

Basic Modelling of Dynamic Engineering Systems

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Basic Modelling of Dynamic Engineering Systems

Uploaded by

Copyright:

Available Formats

Chapter 1

Basic Modelling of Dynamic

1. Experimental modelling. The use of experimental data, obtained by measuring

The steps to be followed in experimental model formulation in general require a considerable

• Formulation of a theory (in our language, a model)

• Experimentation (model (in)validation) and prediction of new observations (model

1.2 The Formal Steps of the Modelling Process

Figure 1.1: The system concept requiring r independent model equations

2. Deﬁne system boundaries and interconnecting variables. The process of

5. Decomposition into subsystems. In many cases, a modeller’s intuition can only

6. Formulation of conservation laws and other mathematical equations. The

7. Mathematical simpliﬁcations and approximations. The mathematical model

8. Numerical solution of the model equations, simulation. Further conﬁdence

Table 1.1: Basic decisions in the modelling procedure

The main steps of the approach are listed in table 1.1.

1.3 Formulation of Conservation Laws for Process Models

1.3.1 Conservation of mass

Figure 1.2: System boundary for macroscopic mass balance

Figure 1.3: Fluid buﬀering vessel

1.3.2 Example: Mass accumulation in vessel

φi (t) initial h(0)

Figure 1.5: Internal structure of model for ﬂuid buﬀering vessel

5. Conservation law: Mass balance

6. Model satisﬁes requirements ? In retrospect, at this stage it is useful to re-

pi (t) = ρgh(t) (1.8)

φi (t) initial h(0)

Figure 1.8: Internal structure of model of vessel and valve

−3 valve area level

−3 flow out valve inlet pressure

Figure 1.9: Response of ﬂuid buﬀering vessel on step in valve opening

0 5000 10000 15000

A = 1.0 m2 ρ = 980 kg/m3

Figure 1.12: Process vessel under ideal mixing assumption

1.3.3 Conservation laws for energy

1.3.4 Example: Energy accumulation in vessel

inlet temperature inlet temperature

outlet temperature outlet temperature

1.4 Microscopic Conservation Laws

For energy, a possible form is

• A microscopic balance such as 1.13 contains three diﬀerent types of terms:

1. Variations per time unit of accumulated energy

∂cA ∂cA ∂cA ∂cA ∂ cA ∂ 2 cA ∂ 2 cA

1.4.1 One-dimensional microscopic balance equations

∂T (x, t) ∂T (x, t) ∂ 2 T (x, t)

with initial- and boundary conditions

T (x, t0 ) = Tinit (x) (1.20)

1.5 Linearity and Linearization of Nonlinear Models

1.5.1 Deﬁnition of linearity

x(t) = x∗ + ∆x(t) u(t) = u∗ + ∆u(t) (1.28)

Because f (x∗ , u∗ ) satisﬁes (1.27), we have

1.5.3 Example: Process dynamics of mixing process

Figure 1.16: Mixing vessel having natural outﬂow through oriﬁce

inlet flow phi2 outlet concentration

outlet flow phi level

Figure 1.17: Responses of mixing process on inlet ﬂow step

inlet flow phi2 outlet concentration

outlet flow phi level

mass balance (component):

where k is a constant, determined by the opening of the oriﬁce and by material