Rigorous Distillation

Dynamics Simulations
Using a Computer Algebra
ZAKIA NASRI, HOUSAM BINOUS
National Institute of Applied Sciences and Technology, BP 676 Centre Urbain Nord, 1080 Tunis, Tunisia

Received 4 June 2009; accepted 26 August 2009

ABSTRACT: In the present paper, Mathematica1, the famous computer algebra, is used to perform steady-
state and dynamic simulations of a multi-component distillation column. In order to describe the vapor
liquid
equilibrium (VLE) relationships and to compute the vapor- and liquid-phase enthalpies, the Soave

Redlich
Kwong equation of state (SRK EOS) is chosen. Rigorous modeling is performed where both mass and
energy balance equations are solved. The Francis weir formula is selected to model column hydraulics. All steady-
state results of the column simulations are found to agree very well with those obtained using HYSYS. Finally,
some insight into column control is given through the application of both the slope and sensitivity criteria. The
computer code, developed at the National Institute of Applied Sciences and Technology (INSAT) in Tunis by the
authors, has served to teach several aspects of chemical engineering to senior undergraduate students at INSAT.
Indeed, students gain from such code a good understanding of energy and mass balances, column dynamic and
control, equations of state and VLE as well as some applied mathematics such as nonlinear algebraic equations,
differential-algebraic systems of equations, and how they can be solved using computer algebra such as
Mathematica1. In addition, they learn about major process simulators such as HYSYS. ß 2009 Wiley Periodicals,
Inc. Comput Appl Eng Educ; Published online in Wiley InterScience (www.interscience.wiley.com); DOI 10.1002/
cae.20385

Keywords: distillation; Mathematica1; applied thermodynamics; SRK EOS; simulation dynamic and control

INTRODUCTION study, the authors humbly attempt to unveil some of the secrets
about such rigorous multi-component distillation simulations
Effluents from chemical reactors need to be separated in order to using a computer algebra that is widely available in universities
achieve product purity commercial specifications. Several unit around the world and a simple quaternary example. In addition to
operations are available for this purpose such as adsorption, steady-state simulations, the dynamic behavior of a distillation
absorption, and stripping, liquid
liquid extraction, leaching, and column was studied in the present work. Indeed, in order to model
last but not least distillation. More than 40,000 distillation units column start-up and shut-down operations, one has to perform
operate in the United States of America alone. This makes the dynamic simulations. Also, a column can be subject to distur-
study of such ubiquitous separation method very important in any bances, in the feed composition or feed flow rate, for example, a
good standing chemical engineering undergraduate curriculum. situation that requires the good grasp of basic ideas about column
Usually, such teaching is based on presenting the McCabe and dynamic and control. Process dynamic and control is usually
Thiele method for binary mixtures and some short-cut methods taught in a separate course. Distillation columns are generally a
for multi-component mixtures. Rigorous simulations are usually good example of a process unit to be selected in order to particle
performed using commercial process simulators such as ASPEN such techniques.
PLUS and HYSYS by Aspen Technology, Inc. (http://www.as- In the first section, we present the fluid package used, which
pentech.com/). Thus students often fail to see the details behind is the Soave
Redlich
Kwong (SRK) Equation of State (EOS).
such calculations and the level of complexity of the problems at The system of DAEs that must be solved in order to obtain
hand (several hundreds of differential-algebraic equations information about the column’s temperature, composition, etc. is
(DAEs) sometimes of index superior to one). In the present given in The Mass and Energy Balance Equations Section. Then,
the Francis weir formula is given. This formula allows one to
model the column hydraulics. We perform steady-state simu-
Correspondence to H. Binous ([email protected]). lations and compare our results with those obtained with
ß 2009 Wiley Periodicals Inc. HYSYS 3.2, a major process simulator by Aspen Technology,

1
2 NASRI AND BINOUS

Inc. (http://www.aspentech.com/), in Steady-State Simulation where

Results Section. Then, the authors present the results of the   
Bi A 2Ai0:5 Bi
dynamic simulations of the distillation column by making step fvi ¼ exp ðZv
1Þ
lnðZv
BÞ


B B A 0:5 B
changes in reflux ratio and rebolier heat duty. In the last section,   ð12Þ
we apply the slope and sensitivity criteria in order to gain some Zv þ B
ln
insights into how one should control the column. Finally, in the Zv
conclusion, the authors share their experience teaching chemical
A similar expression is obtained for the liquid-phase
engineering problems with the help of computer software such as
fugacity coefficient, fli , by replacing the gas-phase compressi-
Mathematica1 and HYSYS.
bility factor, Zv with its liquid-phase counterpart, Zl. These two
compressibility factors are the largest and smallest roots of
Equation (6), respectively.
THE SOAVE
REDLICH
KWONG EQUATION
The liquid and vapor mole fractions are related by
OF STATE
yi ¼ Ki xi with i ¼ 1 to Nc ð13Þ
For a pure component, the SRK EOS [1
5] is given by
where Ki is the equilibrium constant.
RT a One also needs the following expression for the departure
P¼
ð1Þ
ðV
bÞ VðV þ bÞ function from ideality for the enthalpy in order to compute feed
enthalpy as well as the liquid and vapor enthalpy in all stages [4]:
where  
1 ZþB
RTc H D ¼ RTðZ
1Þ þ ln
b ¼ 0:08664 ð2Þ BðRT=PÞ Z
Pc " ! # ð14Þ
d ðRTÞ2 ðRTÞ2
T A
A
ðRTc Þ2 pffiffiffiffiffi dT P P
a ¼ 0:42748 ð1 þ mð1
Tr ÞÞ2 ð3Þ
Pc
where the compressibility factor is equal to: (i) Zv for vapor
T enthalpy, (ii) Zl for liquid enthalpy, and (iii) Zf for the feed, which
Tr ¼ ð4Þ is considered in the present study as saturated liquid.
Tc
It is the authors’ experience that all computations relative to
and the SRK EOS can be performed using Mathematica1 [6].
m ¼ 0:480 þ 1:574o
0:176o2 ð5Þ
The SRK EOS is part of a family of equations called cubic
because the compressibility factor, Z ¼ PV/RT, is a solution of a THE MASS AND ENERGY BALANCE EQUATIONS
cubic equation. This equation, written for a multi-component
mixture where we have used the mixing and combining rules, is In the present section, we give the governing equations that allow
as follows: the steady-state and dynamic simulations to be performed. They
are composed of the total and partial mass balances and the
Z 3
Z 2 þ ZðA
B
B2 Þ
AB ¼ 0 ð6Þ energy balance around the feed stage, the partial reboiler, the total
condenser, and any tray other than the feed tray. These balance
where
equations, written separately for all the column sections men-
C X
X C X
C X
C tioned above (see Fig. 1), are the following:
A¼ yi yj Aij or xi xj Aij ð7Þ
i¼1 j¼1 i¼1 j¼1
(1) Feed tray (k ¼ f):
0:5
Aij ¼ ðAi Aj Þ ð1
kij Þ ð8Þ dMf
¼ F þ Lf
1 þ Vf þ1
Lf
Vf ð15Þ
dt
X
C X
C
B¼ yi Bi or xi Bi ð9Þ dðMf xf ;i Þ
¼ Fzf ;i þ Lf
1 xf
1;i þ Vf þ1 yf þ1;i
i¼1 i¼1 dt ð16Þ

Lf xf ;i
Vf yf ;i for i ¼ 1; 2; . . . ; Nc
P ri P ri
Ai ¼ 0:42747ai and Bi ¼ 0:08664 ð10Þ
Tr2i Tri dðMf hf Þ
¼ FhF þ Lf
1 hf
1 þ Vf þ1 Hf þ1
Lf hf
Vf Hf
dt
For each component, we define the reduced pressure and
ð17Þ
temperature by Pri ¼ P=Pci and Tri ¼ T=Tci , and ai is given
by the same equation as Equation (3) given for the pure (2) kth tray (k 6¼ f and 1 < k < N):
component case. The binary interaction parameter, kij, is assumed
dMk
to be equal to 0 if not available or taken from HYSYS. The ¼ Lk
1 þ Vkþ1
Lk
Vk ð18Þ
equilibrium constants are obtained using the f
f method as dt
follows:
dðMk xk;i Þ
¼ Lk
1 xk
1;i þ Vkþ1 ykþ1;i
f dt ð19Þ
Ki ¼ li for i ¼ 1 to Nc ð11Þ
fvi
Lk xk;i
Vk yk;i for i ¼ 1; 2; . . . ; Nc
 RIGOROUS DISTILLATION DYNAMICS SIMULATIONS 3

problems [8] despite the fact that the dynamics of the energy
equations are usually very fast.

THE FRANCIS WEIR FORMULA

The momentum balance for each tray is neglected. The Francis

weir formula is assumed and provides additional equations.
Indeed, for each stage k, we have a relationship between the tray
liquid hold-up, Mk, and the liquid molar flow rate leaving the kth
stage, Lk:
" P c !#
rk;l Lk Ni¼1 xk;i Mw;i
Mk ¼ PNc Aa hw þ 1:41 pffiffiffi ð27Þ
i¼1 xk;i Mw;i
rk;l l g

where
PNc
P i¼1 xk;i Mw;i
rk;l ¼ ð28Þ
Zk;l RT
For the total condenser and partial reboiler, we make the
following assumption concerning their volume, V0:
Zk;l RT
V0 ¼ 0:5 m3 ¼ Mk for k ¼ 1 or N ð29Þ
P
The Francis formula is only used in the dynamic simulation
where the liquid molar hold-up in all trays must be determined.
Figure 1 Schematic of a distillation column.

dðMk hk Þ STEADY-STATE SIMULATION RESULTS

¼ Lk
1 hk
1 þ Vkþ1 Hkþ1
Lk hk
Vk Hk ð20Þ
dt
Consider a multi-component mixture to be fed to a distillation
column containing N ¼ 27 theoretical stages, a total condenser,
(3) Reflux drum (k ¼ 1): and a partial reboiler and operating at a pressure of 16.212 bar. We
dM1 assume that (i) the pressure is constant on all trays. A rigorous
¼ V2
ðL1 þ DÞ ð21Þ simulation, where variable pressure in the column is considered,
dt
using Fortran 77, is available in the book by Luyben [9]. Another
dðM1 x1;i Þ treatment where variable pressure is considered is given by Choe
¼ V2 y2;i
ðL1 þ DÞxD;i for i ¼ 1; 2; . . . ; Nc
dt and Luyben [7], (ii) the vapor and liquid leaving each tray are in
ð22Þ equilibrium (i.e., tray efficiency is 100%), and (iii) the column is
adiabatic. The column feed, a saturated liquid, is composed of
dðM1 h1 Þ 2.5% ethane, 35% propane, 60% n-butane, and 2.5% n-pentane at
¼ V2 H2
ðL1 þ DÞh1
Q_ C ð23Þ T ¼ 347.8 K. Feed flow rate is set equal to 300 kmol/h. The feed is
dt
entered at the 13th stage. The reflux ratio and reboiler heat duty
are chosen equal to 3.073 and 1.932  106 kJ/h, respectively.
(4) Reboiler (k ¼ N): Composition profiles in the column are given in Figure 2a,b. We
dMN find perfect agreement between our results and the data provided
¼ LN
1
LN
VN ð24Þ by the HYSYS simulation. The temperature profile, equilibrium
dt
constants, liquid and vapor flow rates, and liquid and vapor
dðMN xN;i Þ molecular weights, obtained thanks to the computer algebra
¼ LN
1 xN
1;i
LN xN;i
VN yN;i for i ¼ 1; 2; . . . ; Nc Mathematica1, are plotted in Figures 3
6. All these results are in
dt
ð25Þ agreement with what one gets with HYSYS. It is found that vapor
and liquid molar flow rates are not constant in both the stripping
dðMN hN Þ and rectifying sections of the column. This represents a
¼ LN
1 hN
1
LN hN
VN HN þ Q_ B ð26Þ justification for making rigorous calculation taking both mass
dt
and energy balances into account and not assuming the usual
In the equations above, we have made the usual assumption constant molar overflow (CMO) hypothesis. Liquid density in the
where vapor molar hold-up is neglected, which may give column is given in Figure 7a,b. The reader should note that he
erroneous results for very high column pressures [7]. In addition, must specify to HYSYS ‘‘Use EOS Density’’ in order to get
for the energy equation, we used the full differential equation results in agreement with the present study using Mathematica1.
(i.e., we did not set the LHS of the energy balance equations equal It is found that the distillate bubble-point temperature is equal to
to 0) rather than an algebraic equation, which neglect the time 301.21 K, which allows the use of cold water utility in the
derivative. Indeed, this second approach may cause certain condenser and justifies a posteriori the use of a column pressure
4 NASRI AND BINOUS

Figure 2 Mole fractions of ethane, propane, n-butane, and n-pentane versus stage number: (a) present
work and (b) HYSYS simulation. [Color figure can be viewed in the online issue, which is available at
www.interscience.wiley.com.]

Figure 3 Temperature profile in the column. [Color figure can be Figure 4 Equilibrium constants for ethane, propane, n-butane,
viewed in the online issue, which is available at www.inter- and n-pentane. [Color figure can be viewed in the online issue,
science.wiley.com.] which is available at www.interscience.wiley.com.]
 RIGOROUS DISTILLATION DYNAMICS SIMULATIONS 5

Figure 5 Vapor and liquid flow rates in kmol/h. [Color figure can
be viewed in the online issue, which is available at www.
interscience.wiley.com.]

of 16.212 bar. The condenser heat duty is determined to be equal

to 1.77363  106 kJ/h. The bottom and distillate rates are equal
to 268.17 and 31.83 kmol/h, respectively. The Mathematica1
built-in command used to solve the system of nonlinear algebraic
equations is FindRoot. Here, we have solved a system of 297
nonlinear algebraic equations in 1.81 s using an Intel1 CoreTM
2 DUO CPU T8300 at 2.4 GHz with 3 Gb of memory.

DYNAMIC SIMULATION RESULTS

The dynamic simulation requires solving a system of 325 DAEs

using the built-in Mathematica1 command, NDSolve. When we
Figure 7 Liquid density versus stage number: (a) present work
apply a step change in the reflux ratio or reboiler heat duty, it is
and (b) HYSYS simulation.
important that the initial conditions be a solution of the steady-
state simulation unless NDSolve may adjust the initial conditions,
so that these latter give zero residual for the system of DAEs, or and bottom, respectively. Figures 11 and 12 are a representation
worse NDSolve may fail in finding a solution to the system of of the molar fractions of all components versus time for a step
DAEs. We performed in this study both a 5% step in the reflux change in the reboiler heat duty for the distillate and bottom,
ratio at t ¼ 10 h (see Fig. 8) and a 5% step in the reboiler heat duty respectively. Finally, one can easily study the response of the
at t ¼ 10 h. The dynamic simulation results are obtained in less dynamic system to a perturbation in the feed flow rate. A 5% step
than 12.37 s using an Intel1 CoreTM 2 DUO CPU T8300 at in the feed flow rate gives the results shown in Figures 13 and 14.
2.4 GHz with 3 Gb of memory. In all the cases studied above, we observe that the new
Figures 9 and 10 show the molar fractions of all components steady state, corresponding to the new values set for the reflux
versus time for a step change in the reflux ratio for the distillate ratio and heat duty or feed low rate, is reached after a transient
period of approximately 1 h, which starts at t ¼ 10 h. It is easy to
take advantage the fact that the computations are made using

Figure 6 Vapor and liquid molecular weight versus stage

number. [Color figure can be viewed in the online issue, which
is available at www.interscience.wiley.com.] Figure 8 Reflux ratio versus time (a 5% step).
6 NASRI AND BINOUS

Figure 9 Mole fractions versus time in the distillate for a step in the reflux ratio: (a) ethane, (b) propane,
(c) n-butane, and (d) n-pentane.

Figure 10 Mole fractions versus time in the bottom for a step in the reflux ratio: (a) ethane, (b) propane,
(c) n-butane, and (d) n-pentane.
 RIGOROUS DISTILLATION DYNAMICS SIMULATIONS 7

Figure 11 Mole fractions versus time in the distillate for a step in the reboiler heat duty: : (a) ethane,
(b) propane, (c) n-butane, and (d) n-pentane.

Figure 12 Mole fractions versus time in the bottom for a step in the reboiler heat duty: : (a) ethane,
(b) propane, (c) n-butane, and (d) n-pentane.
8 NASRI AND BINOUS

Figure 13 Mole fractions versus time in the distillate for a step in the feed flow rate: : (a) ethane,
(b) propane, (c) n-butane, and (d) n-pentane.

Figure 14 Mole fractions versus time in the bottom for a step in the feed flow rate: : (a) ethane,
(b) propane, (c) n-butane, and (d) n-pentane.
 RIGOROUS DISTILLATION DYNAMICS SIMULATIONS 9

Figure 15 Ethane mole fraction in the distillate for a 5% step in

the reflux ratio at t ¼ 0. Figure 17 Application of the slope criterion to the distillation
column.

Mathematica1 in order to find the transfer function. These temperature from tray to stray. Indeed, large changes in
functions were obtained using two built-in functions of temperature from stage to stage are indicative of a section of
Mathematica1: FindFit and LaplaceTransform. They are given, the column where mole fractions of important components are
for instance, for the ethane mole fraction in the distillate and changing. Thus, one must maintain temperature at this column
bottom (GD and GB, respectively), for a 5% step in the reflux ratio position. This method is called the slope criterion [11]. We have
at t ¼ 0 h, by computed the slope of the temperature profile given in Figure 3.
Result is shown in Figure 17 and clearly indicates that stage 10
e
0:05s ð1:0279 þ ð6:8874 þ 1:3879sÞe0:05s Þ
GD ¼ ð30Þ and stages near the reboiler and the condenser as candidate
4:9621 þ s locations where temperature should be controlled.
and In order to apply the sensitivity criterion [11], we apply a 7.5%
step change in the reflux ratio, which is one of the two
0:0173e
0:05s ð1:0341 þ ð8:66675 þ sÞe0:05s Þ manipulated variables (i.e., the reflux ratio and the reboiler heat
GB ¼ ð31Þ
8:6675 þ s duty). We look for the locations where there are the largest
Figures 15 and 16 show the dynamic response of the ethane changes in temperature. Results, shown in Figure 18, are in
mole fraction in the distillate and bottom using the transfer agreement with those found using the slope criterion. They
function, shown in red, given above as well as the data points, clearly indicate that stage 10 and the condenser as candidate
shown as blue dots, from the dynamic simulation. One can readily locations for temperature control.
apply such transfer functions in order to control the distillation
column using the well-known Wood and Berry [10] method and
SIMULINK1. CONCLUSION

We have shown how one can successfully apply computer algebra

THE SLOPE AND SENSITIVITY CRITERIA such as Mathematica1 to perform static and dynamic simu-
lations. Relatively large systems of nonlinear algebraic equations
In order to perform good control of a distillation column, one and DAEs can be solved very rapidly thanks to the built-
must select the trays where there are the largest changes in in commands of Mathematica1, FindRoot, and NDSolve,

Figure 16 Ethane mole fraction in the bottom for a 5% step in Figure 18 Application of the sensitivity criterion to the dis-
the reflux ratio at t ¼ 0. tillation column.
10 NASRI AND BINOUS

respectively. The steady-state and dynamic results were com- x liquid composition (mole fraction)
pared to those obtained using HYSYS. Then, they were used to y vapor composition (mole fraction)
perform slope and sensitivity analyses in order to gain insight Z compressibility factor
regarding column control. All computer codes are available, in
the form of a Mathematica1 notebook, upon request from the
corresponding author. These codes are among a series of Greek Letters
computer programs, developed by Dr. Housam Binous at the
fl, fv fugacity coefficients
National Institute of Applied Sciences and Technology (INSAT)
o acentric factor
in Tunis, in order to help his senior and junior students gain
r density (kg/m3)
insight into unit operation problems such as distillation,
absorption, leaching, liquid
liquid extraction, etc. In addition,
the nontrivial nature of the mathematical problem at hand (i.e., Subscripts and Superscripts
large complex systems of DAEs and nonlinear algebraic
equations) makes these codes useful to the applied mathematics i component number
course taught by the corresponding author at INSAT. Finally, a N total number of stages including the condenser and the
new computer laboratory, where Mathematica1 and MATLAB1 reboiler
are taught simultaneously to senior students, has also profited k stage number
from these recent developments and new way of teaching science. v vapor phase
l liquid phase
NOMENCLATURE

Aa active area of the stage (m2) REFERENCES

B bottom flow rate (kmol/h)
D distillate flow rate (kmol/h) [1] G. Soave, Equilibrium constants from a modified Redlich
Kwong
d column inside diameter (m) equation of state, Chem Eng Sci 27(1972), 1197
1203.
F feed flow rate (kmol/h) [2] J. W. Tester and M. Modell, Thermodynamics and its applications,
g 9.81 (m2/s) 3rd edition, Prentice Hall, Upper Saddle River, NJ 1996.
[3] J. M. Prausnitz, R. N. Lichtenthaler, and E. G. deAzevedo,
H vapor enthalpy (kJ/kmol)
Molecular thermodynamics of fluid-phase equilibria, 3rd edition,
hw weir height (m)
Prentice-Hall, Englewood Cliffs, NJ, 1998.
h liquid enthalpy (kJ/kmol) [4] S. I. Sandler, Chemical and engineering thermodynamics, 3rd
Ki equilibrium constant edition, Wiley, NY, 1999.
L liquid molar flow rate (kmol/h) [5] E. L. Henley and J. D. Seader, Equilibrium-stage separation
l weir length (m) operations in chemical engineering, Wiley, NY, 1981.
Mk liquid hold-up (kmol) [6] Z. Nasri and H. Binous, Applications of the Soave
Redlich

Mw molecular weight (kg/kmol) Kwong equation of state using Mathematica1, J Chem Eng Jpn
N number of trays in the column 40(2007), 534
538.
Nc number of components in the mixture [7] Y. S. Choe and W. L. Luyben, Rigorous dynamic models of
distillation columns, Ind Eng Chem Res 26(1987), 2158
2161.
P column pressure (bar)
[8] C. Fuentes and W. L. Luyben, Comparison of energy models for
Pc,i critical pressure (bar)
distillation columns, Ind Eng Chem Res 21(1982), 323
325.
Pr,i reduced pressure [9] W. L. Luyben, Process modeling, simulation and control for
Q_ B reboiler heat duty (kJ/(kmol h)) chemical engineers, Chemical Engineering Series, 2nd edition,
Q_ C condenser heat duty (kJ/(kmol h)) McGraw-Hill International Editions, NY, 1996.
R reflux ratio [10] R. K. Wood and M. W. Berry, Terminal composition of binary
t time (h) distillation column, Chem Eng Sci 28(1973), 1707
1717.
T temperature (K) [11] W. L. Luyben, Evaluation of criteria for selecting temperature
Tc,i critical temperature (K) control trays in distillation columns, J Process Control 16(2006),
Tr,i reduced temperature 115
134.
V vapor flow rate (kmol/h)

BIOGRAPHIES

Zakia Nasri is a PhD student at the National Housam Binous is a full-time faculty member
Institute of Applied Sciences and Technology at the National Institute of Applied Sciences
in Tunis. She earned a Masters degree and a and Technology in Tunis. He earned a Diplôme
Diplôme d’ingénieur in Industrial Chemistry d’ingénieur in Biotechnology from the Ecole
from the National Institute of Applied Sciences des Mines de Paris and a PhD in Chemical
and Technology in Tunis. Her research inter- Engineering from the University of California
ests are in applied thermodynamics and at Davis. His research interests include the
petroleum engineering. applications of computers in chemical
engineering.