1
Models in Chemical Engineering
a
rhe behaviour of a system in the different engineering disciplines and the sciences is studied either
cpetimentally oF theoretically, Experimental study involves conducting experiments on the actual
stem and characterising its behaviour by making quantitative measurements of the dependent
vriables, ‘The steady state of a multistage distillation column, for example, is determined by
seeasuring the time independent compositions of the liquid and vapour phases in different stages.
The temperature profile of a packed bed reactor or a heat exchanger inthis approach is measured
directly by using thermocouples at fixed locations. This approach yields an accurate and realistic
dca of the system behaviour (should our measurements be free of any errors). It, however, is time
onsuming, expensive and labour intensive.
1.1 MODELLING AND SIMULATION
The theoretical approach for analysing a system consists of two steps: (i) modelling, ang
«> simulation. Modelling is the abstract representation of a physical system by equations. ‘These
ict the different physical interactions in a system and are based on the fundamental laws of
physics, typically the conservation of mass, energy and momentum. The equations accurately represent
model the system behaviour. This approach is useful when the different processes and the
shysical interactions occurring are well characterised. It enables us to quantify the different processes
sccurring in a system and its response 10 different inputs. In a chemical reactor, for example, the
wvrction Kinetics should be well understood to generate a good model, Similarly, an accurate
Gescription of the vapour liguid equilibrium (VLE) characteristic of a multicomponent system is a
cnuat to be able to simulate a distillation column. A detailed exposition of the fundamental principles
modelling can be found in Aris (1978),
‘Simulation entails solving the modelling equations either numerically or analytically. The
solution represents the system behaviour and its performance. This book deals with analytical
methods for solving equations which arise in modelling systems. The model (i.e. the describing
ions) is assumed to be available. Throughout this book we will be interested only in the
sulation of the system behaviour, and not in modelling aspects. The analytical methods developed
re can be applied to solve a wide class of linear problems—linear algebraic equations, linear
inary differential equations and partial differential equations.
“The theoretical approach is inexpensive and effective in studying a system comprehensively
“The advent of high-speed computers has facilitated the numerical solution of models of complex
Systems, This has given a big boost to this approach. ‘The disadvantage here is that mispredictions
ch occur due to incorrect modelling or due to wrong assumptions made in modelling. These
sually arise because of a poor understanding of the processes oecurting in the system. The role2 Mathematical Methods in Chemical Engineering
of simulations in understanding system behaviour in various disciplines is discussed extensively in
Aburdeen (1988).
Simulation can be classified broadly into two categories:
Steady state simulation. This involves studying the time invariaut characteristics of a
system. Chemical plants and various units are operated in a continuous mode, usually under steady
conditions. A steady state operation allows us to control the system efficiently and elegantly at a
desired set-point. The behaviour and performance of such systems is analysed by steady state
simulation,
ii) Dynamic simulation. The evolution of the state of a system with time is studied here. This
finds applications in the study of batch operations and other transient phenomena associated with
start-up and shut-down of plants. The dynamic mode of operation of a plant is sometimes preferred
as it can maximise selectivity in a reactor and productivity in a plant (see Lee et al 1980). A
second example arises in biochemical fermentation processes which are typically characterised by
substrate inhibition. The optimum performance of the reactor here is obtained when the process is
ied out in a fed-batch mode. This is a dynamic mode where the substrate is added to the reactor
over a period of time (see San and Stephanapoulous, 1984)
‘The equations generated in modelling can be either linear or nonlinear, Linear equations
can be solved elegantly by analytical methods. This forms the subject of Chapters 1-8. Here we
discuss solution techniques in a generalised framework using the concept of operators.
Nonlinear equations are solved normally using numerical techniques. In the third section,
analytical methods and semi-analytical methods for nonlinear equations are discussed. Here the
emphasis is not on methods of solution (unlike in the first two sections), but on oblaining insight
imo different kinds of system behaviour that can arise due to nonlinearities. Bifurcation theory is
uused for this purpose. This is a theory which is sufficiently general and can be applied to algebraic
equations at one end of the spectrum and to partial differential equations at the other end. It is based
fon the analysis of the linearised equations, and helps us study how a parameter ean influence
system behaviour.
‘The approach adopted in this book, it is believed, will enable the student to develop a deeper
understanding and appreciation of the fundamental concepts behind the mathematics associated
with a problem. The various concepts are developed intuitively. This will enable the aspiring
engineer, with some mathematical inclination, to readily extend the concepts and methods discussed
in this book to the analysis of the behaviour of complex systems which he might encounter.
‘The remainder of this chapter is devoted to discussing different situations where equations of
varying degrees of complexity arise in modelling of chemical engineering systems. Examples of
‘general engineering systems can be found in Kersten (1969).
1.2 LINEAR EQUATIONS
A system of equations is linear when the dependent variables and their derivatives with respect
to the independent variables occurring in it, are raised to the power of unity. A linear equation
should also not contain the product of the dependent variable and its derivatives.
‘A linear equation such that every term in the equation has the dependent variable or its
derivative is said to be a homogeneous equation. A nonhomogeneous equation is one that has
The first
written as
Here F,
(1962).
approach,
therefore.
hhave been
right-hand)Models in Chemical Engineering
a term independent of the dependent variable. This classification as homogeneous or nonhomogeneous
is relevant only for linear systems. Consider the equations
The first three examples, (I.1a)-(I.le), are nonlinear. The fourth (1.1d) is an example of a linear
homogeneous equation, and the fifth (I.le) an example of a linear nonhomogeneous equation (as
sin «° is independent of the dependent variable “u’),
1.2.1 Linear Algebraic Equations
Such systems normally arise in the modelling of the steady states of lumped parameter systems, ic
systems characterised by variables which are spatially independent. Here we have a system of ‘n”
coupled linear equations in ‘n™ unknowns. These equations can be written compactly as
Au=b a2)
where A is an m Xn mattix, b, w are vectors with m coordinates each. Au denotes the matrix A
multiplying the vector u, This operation will be formally defined later. Here we are interested in
determining ‘uw’, for a given A, D, such that (1.2) is satisfied. This equation is a nonhomogeneous
equation if b # 0 (now the right-hand side bas & which is a vector independent of 1).
We will see a few situations where such equations arise. A typical chemical plant consists of
several interlinked units, The effluent stream from a unit serves as the inlet to the next unit
downstream, These units act as junctions/nodes where different streams combine or separate. The
flowsheet in Fig. 1.1 has nine liquid streams connecting five units. An experimental approac
would involve measuring the nine flow-rates to describe the state of the plant. Neglectin
variations across each stream, the mass balance equations across each node at steady state can be
Fy-Fr= Fy Py-Fa= Fs Fa-Fr= Fo Fit Fy= Fo Fe=Fo- Fe (13a)
Here F, represents the volumetric flow rate of the ith stream (see Reklaitis (1983) and Himmelblau
(1962)). In (1.3a) at least four variables have to be specified or determined experimentally. The
remaining five can then be estimated from the equations (model) which are generated by applying
the principle of conservation of mass to cach unit. This is in contrast to the purely experimental
approach where we would have to measure all nine flow rates. Using the modelling approach can
therefore result in a lower investment in the instrumentation of a process plant. Equations (1.3a)
have been written such that F\, Fs, Foy Fo... are experimentally measured, i.e. they occur on the
-hand side. In vectorial form, this system reads as‘Mathematical Methods in Chemical Engineering
0) fA F
in) f
| Fil = | Fo
| Lal Laer
or Au = b,
‘Fh aceuraey of our predictions using (1.3) can be tested by independenily messiei the
tow dies Fa Fa Fe, Fy, Fe A discrepancy between the measured value and he predicted value
tray ovcur, Two plausible explanations for this could be: () Signin density variations in the
a catgeams exist which we have negleted. (ji) Leaks ae present in the plant network which
swe have considered to be nonexistent.
Fig. 11 ‘Typical process flowshest (R: reactor; M: mixer; Dy, Day Ds: separator units;
F; flow rates)
‘The state of the plant in Fg. 1-1 cannot be described by measuring amy set of TON" ables
“The specification of Fy Fs, Fx, Fie for instance, does not allow us 19 detente the other five
artes pniguely, The first three variables are “dependent” as they satisty {he fourth equation in (13)
‘The problem hence reduces to determining five variables from the TNO four equations.
Toncept of “dependence” has wide applications in sensor placement 8 & chemical pl
soneetadron and. Veverka, 1992). This concept is discussed in detail in Chaplet 2.
A atayotem described above i similar to electrical networks, Where the elesic cone fo
through each branch isthe quantity of interest, Equations ofthe form (1-3) th from an applies
cree Gofrs first law at-a junction (see Del Toro, 1986). This analogy leads us 19 olfModels in Chemical Engineering 5
example where we come across linear algebraic equations. It is based on Ohm's law. The voltage
drop V across a resistor R follows the linear relation
V=iR (Ohm's law)
‘The hydrodynamic analogue of this equation relates the pressure drop to the mean velocity v
for laminar flow in a pipe by
apn (2)
This is the classical Hagen-Poisculle equation (see Gupta and Gupta, 1984). The resistance offered
to the flow (32j4/D*) is proportional to the pipe length and varies inversely as its area. This is
analogous to the situation prevailing in electrical networks, where the electrical resistance has the
same features (Halliday and Resnick, 1991),
Consider a liquid entering into a pipe of length ! and diameter D at a fixed pressure P, The
flow distributes itself into two pipes each of length 1; (J:) and diameter D, (D3), see Fis. 1.2
Fig. 1.2 Flow distribution in a pipe network.
Neglecting pressure losses at the junction, and assuming the flow is laminar in each pipe, the
macroscopic momentum balance and the mass balance at the junction yield
(32m) (
(o)+*|
(2200), (22m). f
po) * (pe |
D?v = DEvy, + 12D} |
Here P, is the pressure at which the fluid leaves the system at the two outlets. The set of three6 ‘Mathematical Methods in Chemical Engineering
equations in (14a) can be solved for v, vy. v; fr a fixed (P; P,). They can be reeast vectorilly
saul Rath
DP
or Au = b.
‘Should the flow be turbulent, the friction factor f is not equal to G4/Rey anymore and is
‘obtained from a nonlinear correlation or the Moody diagram (see Gupta and Gupta, 1984). This
sere the equations (14a) nonlinear and prevents us from recasting the set (1.4) inthe vectorial
form (1.4b) as done above.
eanher example where linear algebraic equations occur isin an idealised extraction unit. A
ccounter-current three-stage extraction unit is shown in Fig. 1.3. The component A is present in
phase E (extract) along with a nondiffusing substance as a binary mixture, tis extracted into
Stage1 __Stage2_— Stages
% L»x, [> % X(P)
yt se Ye)
Fig. 13. A three-stage counter-current extraction unit (X,Y are mole ratios leaving the ih stage)
phase R (raffinate) by a nondiffusing solvent. The streams leaving the ith stags ave denoted by
rreSabseript The mole ratio of A in phase E(R) is denoted by YO. A material balance of
Component A across each extraction stage (see Treybal, 1981) yields
Ey + RXa = RXs + EVs
E.Yy + RX, = Ea + RX
ExYa + RXq = EY + RX
(x) = moles of A/moles of non-A in phase E(R) coming out of stage # (this is a mole ratio and
not # mole-fraction)
F(R, refers to flow rates in terms of moles of non-A in phase F(R) of each stage. THs is
a consignt and does not vary between the different stages as only species A diffuses from one phase
aon ther The assumption of a linear equilibrium relationship (in terms of mole ratios) for the
‘compositions leaving the ith stage yields three more equations
¥;= KX; fori= 1.2.3 (5b)
“These, six equations in eight compositions (Nop Xiv-eX3. Vix Yanue¥s) can be solved by specifying
wae Saables: It is most convenient to specify Y and Xo the inlet compositions of the two phasesModels in Chemical Engineering
The above set of equations can be written vectorially as
RR & 0 0
K 0
or Au = 6.
‘The vector (Xi, Yin Xo Yo. Xs. Ys) can be obtained for a given E,, Ry K. The assumption of
« linear equilibrium relationship between ¥; and X, (1.5b) could be invalid. This can give rise to
4 discrepaney in the predicted and actual system behaviour.
Systems of equations of the form (1.2) also arise while solving linear elliptic partial differential
equations (see Chapter 5) by finite differences. In such situations the dimension of the system. i.e
number of equations ‘n', ean be very large.
1.2.2 Ordinary Differential Equations (Initial-Value Problems)
This class of equations describe the dependence of different quantities on one independent variable.
Such problems usually occur in modelling the transient behaviour of spatially homogeneous systemas
‘The independent variable here is time. Alternatively, in a steady state problem, this could be u
spatial coordinate. Equations of this kind can be cast in the vectorial form
i =Autb (1.6a)
where the dot over w represents the derivative with respect to the independent variable. This
uation represents a system of coupled first order linear ordinary differential equations ax long as
» is independent of u. In (1.6a), “b” is a nonhomogeneity. In a transient problem (when the
independent variable is time), the evolution of each of the coordinates or elements of wu with time
is to be determined such that (16a) is satisfied subject to somé conditions on u. In an initial-value
Problem, these conditions specify all the coordinates of w at the same value of the independent
variable, usually time ¢. When the condition is specified at the point where the independent variable
is zero, these take the form,
(0) = (1.60)
Equation (1.60) is called the initial condition of w (since usually the independent variable is time,
and 1 = 0 denotes the initial instant of time. The problem (1.6a)-(1.6b) is called an initial valuc
problem.
A boundary-value problem, on the other hand, has some coordinates of 1 specified at one
instant of time and the remaining ones at another instant of time, The treatment of boundary value
Problems is taken up in Chapters 5-8 and 9-12. In this section we restrict ourselves to initial value
problems,
A co-current shell and tube heat exchanger at steady state is shown in Fig. 1.4, This is an
‘example of a system where the model yields an initial value problem. The hot (cold) stream isMathematical Methods in Chemical Engineering
Fig. 1.4 A co-current shell and tube heat exchanger,
flowing through the tube (shell) side (see Kern (1950) and Holman (1972)). Assuming that
heat transfer between the fluids (across the tube) is governed by an overall heat transfer coefTiel
LU and occurs across the perimeter P, the steady axial variation of temperature of the two st
is given by
ay
PuGuan
= -UPTy ~ Te)
ae
eG de Se = UPC ~ To
Here p. Gy 4 denote the density specific heat and flow-rate ofthe streams. Subserips Hy C dea
see es cra scams, respectively, These equations are subject 10 initial conditions which sped
Tha Te at c= 0, the entrance point (since the heat exchanger is co-current.
Ty(z = 0) = Two» Te 0) = Too
‘The system (1.7a) cam be recast in vectorial form as
) [awe
PuGuda — PuGuan
Ue. _ SUE
[PeGeqe — PeGede
Tw
or i= Au.
“The independent variable in this problem is “z", the axial coordinate, This is an example
«a homogencous equation as it has no term independent of u, i.e. b = 0 in (1.62). This is st
state model as Ty, Te are independent of time “7”. The two conditions (1.7) are both specifies
P= 0, The system (1.8) is hence mathematically an initial-value problem, though the indepem
‘variable is not time there.
"Another example where a similar system of equations is encountered is in the modell
a batch reactor (Fig, 1.5) sustaining a network of single-phase first order reactions:
opine
Fa
D
‘The reactor is assumed to be isothermal. The rate constants k, can then be treated as consModels in Chemical Engineering
Fig. 1.3 A well-stirred batch reactor sustaining a network of reactions,
The time variation of the concentration C, of the ith species in the reactor is given by the mass
balance equations (see Fogler, 1992) as
of the reactor
Cit =0) = Cue Ct = 0) = Com Ct = 0) = Com, Cot = 0)
Equaiion (1.9a) can be recast in vectorial form as
0
lc
le
0) Cc
0
0
0
°
le ollee
aepeCre ME = UPC, ~ Te) - UA Te ~T)10 Mathematical Methods in Chemical Engineering
where Uj. Pris the overall heat transfer coefficient and perimeter of the shel surface respectively
In vectorial form the system (1.10) now becomes modified as
ra [= UP_ Irn 7
Hl | PxGudn Punta = ely, (lla)
é UP. - UP - UR || 7 AT,
cl pepe | PcG | a
or a= Aut cute)
‘This modified system is clearly nonhomogeneous, as here the term “bof (16a) is nonzero.
Tee edtrtion ofa reztant undergoing a first order reaction A ~» B in an ideal isothermal
continua stred tank reactor (CSTR), See Fig 1.6, is governed by the mass balance equation
ICL Cy ~ Ca) ~ ECA a1)
where
V = reactor volume
4g = volumetric flow rate of reactant stream
Cy = concentration of feed stream
k= reaction rate constant
Cy
ll
Cy
Fig. 1.6 A well-strred continuous tank reactor.
‘The term qCy is independent of C4, the dependent variable. This renders the system (1.12)
nonhomogeneous.
‘Systems of the form (1.6) also arise while solving linear parabolic partial differential equations
(sce Chapter 5) using finite differences. The variables are discetised in the spatial direction. The
Spstem behaviour now is represented by a finite number of variables governed by equations which
can be recast in the form (1.6a).
1.23. Partial Differential Equations
‘These equations occur widely in modelling systems in different areas of chemical engineeringModels in Chemical Engineering u
There is more than one independent variable in such problems. These are also more generically
called distributed parameter systems as now the variables are dependent on spatial coordinates,
Consider a spherical pellet of radius R at a uniform temperature T; dipped in a well-stirred
bath maintained at a constant temperature T,, The pellet temperature varies with time and position
tuntl it reaches the equilibrium value T, uniformly inside. The variation with position occurs due
to the finite value of thermal conductivity. The time and spatial dependence of pellet temperature
T is governed by the energy balance equation
aT gl a(,2 a0)
a 2 ar or)
(a3)
Here, a represents the thermal diffusivity of the pellet (sce Bird et al., 1960). This is a typical
example of a homogeneous linear parabolic partial differential equation.
Consider next a finite sized cylindrical pellet, for example (0