82 | Part I: Introduction

The multilayer ANN architecture is built on the following rules.

– Artificial neurons are organized in layers, each layer consisting in a set of neurons
having the same activation function.
– Each ANN has one input layer, which is the input gate for the data to be processed,
one output layer, which is the gate of delivering the processed data, and several
intermediate layers named hidden layers, where the input data is processed step
by step and sent towards the ANN output.
– The connection paths (represented by arrows) are oriented from the input towards
the output.
– Each neuron of a layer is only interconnected with all neurons of the neighboring
(adjacent) layers.
– Each neuron receives data from all neurons of the previous layer, computes its
output and sends data to all neurons of its next layer.
– The ANN may have a various number of hidden layers and neurons in each hidden
layer, but the number of neurons in the input and output layer are fixed by the par-
ticular modeling problem (the dimension of the input and output pair examples
of training data).

The ANNs may build complex models and are very efficient especially for cases when
the laws governing input and output are not known or insufficiently formalized. They
are trained on the statistical basis of the implicit information contained in the training
examples. ANNs can also be used for one special case of modeling, namely for clas-
sification. ANNs may separate the elements of a set in classes, on the basis of a priori
settled classification categories or by discovering these categories by themselves. The
capacity of this structure of connected computing elements to process data using a
very short computing time, compared to the numerical methods, is a very much ap-
preciated feature for saving computer resources.

Simulation
Copyright © 2014. De Gruyter, Inc.. All rights reserved.

Simulation is the process of imitating reality using a model. The model is solved and
its solutions represent the variations induced in the process by the inputs, or construc-
tion constants. Simulators are used in industry to train the operators before operating
a process and to test different alternatives of design and control.

3.4 Examples of mathematical models

Example 3.9. Non isothermal CSTR

k
Considering the CSTR (Fig. 3.2) in which an exothermic first order reaction A 󳨀→ B
takes place. The heat of reaction is ΔHr < 0. According to the procedures proposed at

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.
 3 Mathematical modeling | 83

the beginning of the chapter,

1. elaborate a clear and correct flow sheet or drawing of the process described; if the
process is too complex, divide it in lower complexity modules;
2. identify all variables in the process and understand the correlation cause-effect
between them: first inside the modules and second between the modules.

The variables of the process are specified on the drawing.

3. Elaborate the lists of variables and constants: output variables, input variables,
design and construction parameters, thermodynamic and kinetic constants.

Output variables are those of interest from the point of view of the reaction: F, CA , CB ,
T∘.
Input variables are those influencing the output ones: Fvi , CAi , Ti∘ , Fvagi , Tagi
∘ .
State variables: V, T ∘ , ρ , c , C , C , T ∘ , T ∘ .
p A B ag w
Properties of the reactants/products, physical constants: ρA , ρB , MA , MB , cpA , cpB , k0 ,
Ea , ΔHr , R, g, ρag , cpag .
Design and construction parameters: AR , Kv0 , Vmin , Vj , KT , AT , Mp , cpp , KTi , KTag , Ap ,
lp , dp , λ .

Nomenclature:
3
F, Fvi – outflow and input flow of the reactor ( ms )
CAi , CA , CB – molar concentrations of the reactant in the input flow and of A and B in
the reactor ( kmol
m3
)
T ∘ – temperature inside the reactor and in the outflow (K)
Ti∘ – temperature inside the reactor and in the outflow (K)
3
Fvagi – cooling agent flow in the jacket ( ms )
∘ , T ∘ – input and jacket temperature of the cooling agent (K)
Tagi ag
T ∘ – temperature of the inner wall of the reactor (K)
w
V, Vmin , Vj – volume of the reaction mass in the reactor, minimum volume allowed in
Copyright © 2014. De Gruyter, Inc.. All rights reserved.

the reactor for normal functioning of the level control and volume of the jacket (m3 )
ρA , ρB , ρ – density of A, B and of the mixture of A and B ( mkg3 )
MA , MB – molar mass of A and B ( kmol m3
)
cpA , cpB , cp – heat capacity of A, B and of the mixture ( kgJ K )
k, k0 – rate constant and pre-exponential factor ( 1s , 1s )
J
Ea – activation energy ( kmol )
J
R – universal gas constant (= 8,314 kmol K
)
g – gravitational acceleration (= 9.8 sm2 )
J
ΔHr – heat of reaction ( kmol )
ρag , cpag – density and heat capacity of the cooling agent ( mkg3 , J
kg K
)
KT , AT – heat transfer coefficient and heat transfer area ( mW2 K , m ) 2

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.
 84 | Part I: Introduction

KTi , KTag – partial heat transfer coefficients inside the reactor and in the jacket ( mW2 K )
Mw , cpw – mass of the reactor wall and its heat capacity (kg, kgJ K )
Ap , lp , dp – cross area of the outflow pipeline, the length of the pipeline, diameter of
the pipeline (m)
λ – coefficient of pressure loss on the pipeline (−)

4. Elaborate the list with simplifying hypotheses:

– the reactor is perfectly mixed (this shows we can consider the same concen-
trations, same density, same temperature at each point of the mass of reaction
and also at the reactor’s output) and all reactions take place in liquid phase
(no vapor phase to be considered);
– the reaction does not continue in the output pipeline (it means that the con-
centrations in the reactor are the same as those at the end of the pipeline);
– the reactor is cylindrical (simplifies the calculus of height of the liquid inside
the reactor);
– the reactor and pipelines are perfectly isolated thermally (this simplifies the
heat balance because it does not consider the heat losses; usually these are
around 5% of the heat load);
– the fluctuations of temperature are not too large, meaning properties do not
change with temperature;
– the jacket is completely filled with cooling agent (this means Fvagi = Fvag );
– the wall of the reactor is considered “thin” from the heat transfer point of view
(this means the transfer through the wall is instantaneous and does not incur
any delay).

5. Write the equations of the model if possible in the “natural” sequence of the devel-
opment of the process; ensure that all unknowns in the equations are expressed
in new “secondary” equations;
Each variable on the lists is assigned one equation; if one variable is not on the
lists, it should be assigned a new equation until all variables are “covered” by
Copyright © 2014. De Gruyter, Inc.. All rights reserved.

equations. Usually, for flows, momentum balance equations, for temperatures,

energy balance equations and for mass and concentrations, mass balance equa-
tions are used.

Thus, for F: either

lp ρ v2 d
ρ ghAp − λ A = (A l ρ v) (3.38)
dp 2 p dt p p
or the control equation
F = F0 + K V0 (V − Vmin ). (3.39)

is assigned. The second equation expressing the flow is the ‘‘level control” equation,
where Kv0 is the gain factor of the level controller. The outflow is increasing directly

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.
 3 Mathematical modeling | 85

with the increase of the volume of the mass of reaction over the minimum value re-
quired for a normal functioning of the control loop.
The first equation gives the solution for v, needing an extra equation for F: F =
vAp . Looking for the “new” variables we notice h is on no list, V, v, and ρ are on the
list of state variables and we should link them to some “new” equations:

V
for h: h = ; this equation introduces the need of another equation for V (3.40)
AR
d
for V: Fvi ρA − Fρ = (Vρ ) (3.41)
dt
for ρ : ρ = MA CA + MB CB (Tab. 3.1) (3.42)

for CA and CB component mass balance equations will be written

d
Fvi CAi − FCA − VkCA = (VCA ) (3.43)
dt
d
0 − FCB + VkCA = (VCB ). (3.44)
dt
Whatever is transformed from A, contributes to B (+VkCA ).
For T ∘ should be written the energy balance equation

d
Fvi ρA cpA Ti∘ − Fρ cp T ∘ − KT AT (T ∘ − Tag
∘
) − VkCA ΔHr = (Vρ cp T ∘ ) (3.45)
dt
for cp : cp = xA cpA + xB cpB (Tab. 3.1) (3.46)
M M
for xA : xA = CA A and xB = CB B (3.47)
ρ ρ
∘ ∘ ∘ d
for Tag : Fvag ρag cpag Tiag − Fvag ρag cpag Tag + KT AT (T ∘ − Tag
∘
)= (V ρ c T ∘ ) (3.48)
dt j ag pag ag
the agent flow is the same since the jacket having the volume Vag is completely full;
in this example, the inner wall of the reactor is considered “thin”, this having as a
consequence the existence of only one KT . If the wall is not thermally “thin” (Fig. 3.11),
Copyright © 2014. De Gruyter, Inc.. All rights reserved.

the equation takes another form:

∘ ∘ d
Fvag ρag cpag Tiag − Fvag ρag cpag Tag + KTag AT (Tw∘ − Tag
∘
)= (V ρ c T ∘ ) (3.49)
dt j ag pag ag
d
KTi AT (T ∘ − Tw∘ ) − KTag AT (Tw∘ − Tag
∘
)= (M c T ∘ ). (3.50)
dt w pw w
A “thick” wall is either thick physically having an important dimension (e.g. the
heat exchanger of the polyethylene produced in the high pressure process has a wall
of stainless steel of 20 cm), or it is thermally “thick” being thin physically, but a very
bad thermal conductor (e.g. an enameled steel wall).

6. Solve the system of equations and interpret the results.

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.
 86 | Part I: Introduction

KTag KTi

T°
T°i
T°w

T°ag

L Fig. 3.11. The wall is thermally “thick”.

Thus, the example of a CSTR involving practically only four variables describing the
production quantity and its quality has a simplified model of 12–14 equations, 6–7 dif-
ferential and 6–7 algebraic. The system is nonlinear and can be solved using a MatLab
subroutine ODE15s, ODE23, ODE45. The interpretation of the results shows the behav-
ior of all variables in time when one or several inputs change or when some of the
characteristics of the process change (e.g. transfer coefficients due to deposits of lime-
stone in the shell of the heat exchanger or in the cooling jacket).

Example 3.10. Cascade of three CSTRs

In some processes, one reactor is not enough and several reactors in cascade are used.
The reactors in the example are isothermal, perfectly mixed, perfectly insulated and
the reaction takes place only in the liquid phase. They function at the same tempera-
ture, have different holdups, same density and then the different outflows. The reac-
k
tion is A 󳨀→ B, with ΔHr = 0.
Copyright © 2014. De Gruyter, Inc.. All rights reserved.

Fi , CAi

V1 V2 V3

F1 , CA1 , CB1 F2 , CA2 , CB2 F3 , CA3 , CB3

Fig. 3.12. Cascade of three CSTR.

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.
 3 Mathematical modeling | 87

The output variables are F3 , CA3 , CB3 .

The input variables are Fi , CAi .
The state variables are V1 , V2 , V3 , CA1 , CB1 , CA2 , CB2 , CA3 , CB3 , F1 , F2 , F3 .
The model has 12 equations (first 9 independent), since the output variables are state
variables as well:
d
Fi ρ − F1 ρ = (V1 ρ ). (3.51)
dt
d
F1 ρ − F2 ρ = (V2 ρ ) (3.52)
dt
d
F 2 ρ − F3 ρ = (V ρ ) (3.53)
dt 3
d
Fi CAi − F1 CA1 − V1 kCA1 = (V C ) (3.54)
dt 1 A1
d
F1 CA1 − F2 CA2 − V2 kCA2 = (V2 CA2 ) (3.55)
dt
d
F2 CA2 − F3 CA3 − V3 kCA3 = (V3 CA3 ) (3.56)
dt
F1 = KV1 (V1 − Vmin1 ) (3.57)
F2 = KV2 (V2 − Vmin2 ) (3.58)
F3 = KV3 (V3 − Vmin3 ) (3.59)
d
0 − F1 CB1 + V1 kCA1 = (V1 CB1 ) (3.60)
dt
d
F1 CB1 − F2 CB2 + V2 kCA2 = (V2 CB2 ) (3.61)
dt
d
F2 CB2 − F3 CB3 + V3 kCA3 = (V C ) (3.62)
dt 3 B3
The system to be solved has thus 12 differential equations and we solve the first nine
equations (3.51)–(3.59) of the equation system for F1 , F2 , F3 , CA1 , CA2 , CA3 , V1 , V2 , V3 .
With these values, CB1 , CB2 , CB3 are calculated. The steady state values are required to
be able to solve the system.
Copyright © 2014. De Gruyter, Inc.. All rights reserved.

Example 3.11. Binary distillation column [15]

A binary distillation column is considered with N trays which separates two products
with the same relative volatility, α , along the column (on each tray, the volatility is the
same).

List of variables:

Output variables: FB , FD , xB , xD .
Input variables: FF , xF .
State variables: Mj , Lj , xj , yj , MB , xB , yB , MD , xD , FR , V, FB , FD .

Each distillation unit (tray) is represented in Fig. 3.15 + section with the weir.

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.
 88 | Part I: Introduction

FD , xD

FF , xF

FB , xB
Fig. 3.13. Binary distillation column.
Copyright © 2014. De Gruyter, Inc.. All rights reserved.

Fig. 3.14. Distillation column Illudest in the

Control Laboratory of the Faculty of Chemistry
and Chemical Engineering, University Babes-
Bolyai, Cluj-Napoca, Romania.

Simplifying hypotheses:
1. Each tray has the same efficiency (Et = 100%).
2. Vapors leaving the tray are in equilibrium with the liquid phase on the tray.

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.
 3 Mathematical modeling | 89

Lj+1 , xj+1
V, yj j+1th tray

Mj , xj jth tray

V, yj–1 j–1th tray Fig. 3.15. The unit of distillation + section with
Lj , xj the weir.

3. The column is fed on the tray NF with the feed flow FF at the boiling tempera-
ture corresponding to the concentration xF of the volatile component in the liquid
phase.
4. Each tray is perfectly mixed and has a holdup Mj with the molar composition xj ,
of the liquid phase and yj of the vapor phase. The holdup of the vapors on the tray
is negligible (the vapor mass is ca. 1000 times smaller than that of the liquid in an
atmospheric column).
5. The holdups of the bottom are MB (molar composition xB and yB ) and that of the
reflux tank MD (molar composition xD and 0 - the condensation is total).
6. The bottom molar flow is FB with its composition xB and the distillate molar flow
is FD , with its composition xD.
7. The reflux molar flow is FR and the vapor molar flow is V.Supposing the values of
the latent heat of vaporization substantially close, it can be approximated that one
mole of most volatile component vaporizes at the cost of condensing of one mole
of less volatile component, having the result that there is no need for an energy
balance equation on the column and the molar vapor flow Vis constant along the
column.
8. On each tray, the internal molar reflux depends on the holdup of the tray and is Lj .
9. The dynamic response of the condenser is much faster than that of the column.
10. There is no chemical reaction on the trays.
Copyright © 2014. De Gruyter, Inc.. All rights reserved.

Having 4 variables on each tray, the bottom with three variables, the distillate drum
with two and 4 other flows to be determined (FR , V, FB , FD ) it means the model has to
have 4N+9 variables.

Tray equations:
d
Lj+1 − Lj = M (3.63)
dt j
d
Lj+1 xj+1 − Lj xj + Vj−1 yj−1 − Vj yj = (M x ) (3.64)
dt j j
α xj
yj = (3.65)
1 + (α − 1)xj

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.
 90 | Part I: Introduction

Lj = f (Mj )
or Francis equation Lj = 1.837lj h3/2
j , (3.66)

where lj is the length of the weir and hj is the height of the liquid over the weir.

Feed tray equations:

d
FF + LNF +1 − LNF = M (3.67)
dt NF
d
FF xF + LNF +1 xNF +1 − LNF xF + VNF −1 yNF −1 − VNF yNF = (M x ). (3.68)
dt NF NF
The equilibrium equation of the feed tray is the same with all tray equilibrium equa-
tions.

Distillate drum equations:

d
V − FD − F R = M (3.69)
dt D
d
VyN − FD xD − FR xD = (M x ). (3.70)
dt D D

Bottom equations:
d
L1 − FB − V = M (3.71)
dt B
d
L1 x1 − VyB − FB xB = (M x ) (3.72)
dt B B
α xB
yB = . (3.73)
1 + (α − 1)xB

The control equations:

Copyright © 2014. De Gruyter, Inc.. All rights reserved.

FD = FD0 + KvD (MD − MDmin ) (3.74)

FR = FR0 −K vR (xD − xDset ). (3.75)

The reflux has to decrease when xD > xDset the setpoint value for xD :

V = V0 + KvV (xB − xBset ). (3.76)

The vapor flow has to increase when xB >xBset .

The total number of equations is 4N + 9 as is the number of variables.

Example 3.12. The efficiency of an esterification reaction (in %) depends on the molar
concentrations (mol/l) of the reactants (acid and alcohol). A factorial experiment at

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.
 3 Mathematical modeling | 91

Table 3.4. Input variables.

Input variables x1 x2

Center level 1.0 2.4

Variation step 0.2 0.4

Table 3.5. Results of experiments.

Input variables (concentration mol/l) Output variable (efficiency %) Output average (%)
x1 x2 yI yII y

1.2 2.8 0.5 0.3 0.4

1.2 2.0 0.8 0.6 0.7
0.8 2.8 0.2 0.1 0.15
0.8 2.0 0.55 0.25 0.40

two levels [9] was developed, in order to determine the correlation between the output
and input variables. The aim is to determine a linear regression y = a0 + a1 x1 + a2 x2 .
There is a repetition of the measurements for each combination; the average is
taken into consideration.

1 1.2 2.8
[1 1.0 1.0 1.0 1.0 4.0 4.0 9.6
[ 1.2 2.0 ]
] T [ ] [ ]
X=[ ] ; X = [ 1.2 1.2 0.8 0.8 ] ; X T X = [ 4.0 4.16 9.6 ] ;
[1 0.8 2.8 ]
[ 2.8 2.0 2.8 2.0 ] [ 3.6 9.6 23.7 ]
[1 0.8 2.0 ]
Δ = det(X T X) = 0.4096 (3.77)

0.68
T −1 T [ ]
A = (X X) (X Y) = [ 0.624 ] , giving a regression y = 0.68 + 0.624x1 − 0.38x2 .
[ −0.38 ]
Copyright © 2014. De Gruyter, Inc.. All rights reserved.

Statistical processing [13]:

(a) Weight of the coefficients in the model.

2
Sy2 ∑nj=1 ∑rl=1 (yjl − yj̄ )
S2 {ai } = =
n n (r − 1)
(0.5 − 0.4)2 + (0.3 − 0.4)2 + (0.8 − 0.7)2 + (0.6 − 0.7)2 + (0.2 − 0.15)2
{ }
+ (0.1 − 0.15)2 + (0.55 − 0.4)2 + (0.25 − 0.4)2
=
4⋅1
= 0.0041, where n = 4 and r = 2. (3.78)

Agachi, P. S., & Cristea, M. V. (2014). Basic process engineering control. ProQuest Ebook Central <a
onclick=window.open('http://ebookcentral.proquest.com','_blank') href='http://ebookcentral.proquest.com' target='_blank' style='cursor: pointer;'>http://ebookcentral.proquest.
Created from uta-ebooks on 2020-11-12 05:38:16.