Fundamental Modeling and Simulation of a Binary

Continuous Distillation Column

H. S. Truong, I. Ismail, R. Razali

Electrical and Electronic Engineering Department
Universiti Teknologi PETRONAS
31750 Tronoh, Perak, Malaysia
hoang_son_truong@yahoo.com, idrisim@petronas.com.my, radzuan_razali@petronas.com.my

Abstract— Distillation columns are important unit operations in multiple outputs (MIMO) structure and the presence of severe
chemical process plants. This paper reviews some techniques of disturbances during operation [1]. The nonlinearity of
modeling and simulation of distillation columns and describes a distillation columns is well known. It has been shown that the
model for a lab-scale binary continuous distillation column. The purer the products get, the more nonlinear the system becomes
focus is on the dynamic behavior of the product compositions [3]. A distillation column is also a typical example for an
under feed disturbances. The simulation results show that the MIMO system in which there are strong interactions between
composition responses to disturbances are close to the response of the variables. The interactions occurring between the inputs
a first order system. The response to change in feed composition and the outputs are difficult to identify. The disturbances to a
has larger gain than the response to change in feed flow rate. The
distillation column can come from many sources. They can
paper also compares the simulation results with some of other
works.
come from the feed (feed flow rate, feed composition), from
the pressure inside the column, from the cooling water etc.
Keywords- Binary Distillation Column, Fundamental These difficulties pose numerous challenging control problems
Modeling, Simulation. and also attract a large number of researchers from different
disciplines.
I. INTRODUCTION
Distillation column is considered one of the most important
unit operations in chemical engineering, and also the most
studied unit operations in terms of control [1],[2]. A distillation
column is used to separate a mixture into its components by the
application and removal of heat. It consumes a huge amount of
energy in both heating and cooling operations. There are many
types of distillation columns based on different classifications
such as: batch, continuous, binary, multiproduct, tray, packed.
In this paper we focus on continuous binary distillation
columns since continuous columns are dominant in industry
and binary columns are usually referred to as a foundation by
the researchers when they examine other types of distillation
columns.
A simple two-product continuous distillation column is
shown in Fig. 1. The column has N stages on which the vapor-
liquid equilibriums occur. The feed enters the column on the
stage N F . This stage divides the column into a rectifying
section and a stripping section. Near the bottom of the column
is a reboiler which provides energy to the column. The mixture
is heated to form a flow of vapor rising up inside the column.
In the stripping section, the less volatile component is enriched
while in the rectifying section the more volatile component is
enriched. The top product is condensed by the condenser from
which there is a reflux flow back to the top of the column to
enhance the purity of the product.
The difficulties in controlling distillation columns lie in
their highly nonlinear characteristics; their multiple inputs Figure 1. A Two-product Distillation Column

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 18:06:43 UTC from IEEE Xplore. Restrictions apply.
 In order to control a distillation column, the first step is to B. Wittgens and S. Skogestad carried out an evaluation of
develop a model of the column. By modeling we can dynamic models of distillation columns with emphasis on the
understand the behavior of the column, predict future reaction initial response [5]. They found out that the most important
and therefore devise a control structure for the column. This parameters are the liquid holdup, the liquid hydraulic time
paper reviews some important techniques in distillation column constant and the vapor constant that represent the initial effect
modeling and then describes a model for a lab-scale binary of a change in vapor flow on liquid flow.
continuous distillation column.
Abdulla et al. [1] have done a quite complete review on the
The paper is organized as follows. A review of modeling of recent nonlinear modeling applications in continuous
continuous distillations is presented in part II, the modeling of distillation column. The summary states that the empirical
our distillation column is detailed in part III, the simulation modeling has been preferred in industry because of its
results are shown in part IV, and the conclusion and future simplicity compared to the fundamental model; and the current
work are presented in part V of this paper. development focuses on hybrid models, which can exploit the
advantages of both fundamental model and empirical model;
II. MODELING OF CONTINUOUS DISTILLATION COLUMNS and that the neural network method is used the most to
combine with the fundamental model in empirical modeling.
Modeling of distillation columns is often classified into
three groups: fundamental modeling, empirical modeling and In the case of fundamental modeling, the model is often
hybrid modeling [1]. In fundamental modeling, the model is simulated to understand the column’s dynamic behavior. The
constructed based on the physical properties of the system, development of distillation column’s simulation has been going
such as the preservation of mass, energy and momentum. And along with the growth of computing capacity. As of 1930s and
depending on the levels of accuracy of the assumptions we will 1940s only graphical methods and simple short-cut models
have different models ranging from simple to rigorous models. were used to get insights of the steady-state behavior of the
This method of modeling has the advantage of global validity, distillation columns [6]. The fast-growing of computing power
accuracy and it gives more complete process understanding. has allowed the use of more complex and rigorous models.
However, this method is quite complex for controller design Computer programming and the numerical methods to solve
with huge amount of computation and simplifications are often the differential equations play an important role.
needed [1]. The empirical modeling (sometimes called the
black-box modeling) utilizes the input and output data from the III. MODELING OF THE APC
operation of the column to build the relationship between the
input and the output. With this method we do not need to The APC (Advanced Process Control) column is a pilot
understand the inner dynamics of the column, and the distillation column that has 15 trays and equipped with a DCS
computation can be reduced. But in using this method we have control system. The feed is positioned at tray 7. The model is
to carry out experiments on the real column, and the results developed based on a model by [7]. In the model the following
may not be applied for other column, even the results from one assumptions are made:
column can be different if the column’s conditions are different 1. Binary mixture, the feed contains only two components
between the experiment and the actual operation of the column. 2. The pressure inside the column is fixed by controlling
The hybrid modeling (or the ‘grey-box’ model) combines the the cooling water
fundamental modeling and the empirical modeling. This 3. Constant relative volatility, α = 1.5
method utilizes the advantages of the other two, but in order to 4. Constant molar flows
do that we need a well-structured model in which we have to 5. No vapor holdup, the vapor holdup on each tray is
decide which part of the model to use fundamental technique negligible
and which part to use empirical data. In our project we will 6. Linear liquid dynamics
focus mainly on the fundamental modeling, even though the 7. Equilibrium on all stages
empirical model is dominantly used in the industry. The reason 8. Total condenser, there is no vapor holdup in the
is that we want to understand the dynamics of the distillation condenser
columns, and since the black-box model “may not be used to
predict the behavior of the system at other operating The total material balance equation on stage i is:
conditions” [4].
S. Skogestad [2] did a critical survey of literature on d (1)
M i = Li +1 − Li + Vi −1 − Vi
dynamics and control of distillation columns up until 1991. The dt
paper summarized the simplifications of the rigorous model where Mi is the liquid holdup on tray i, Li and Vi are the liquid
since no references had been found on solving all the equations flow rate and vapor flow rate that come towards tray i.
of the rigorous model. The simplifications are aimed to the The material balance for the light component on tray i is:
vapor dynamics, to the energy balance and to the liquid flow d
dynamics. The paper recommends not neglecting liquid (M i xi ) = Li +1 xi +1 − Li xi + Vi −1 y i −1 − Vi y i (2)
dynamics (i.e. not assuming constant liquid holdups) due to the dt
fact that the initial response, an important factor in feedback where xi and yi is the composition of the light component and
control, is largely affected by the liquid holdups. heavy component on tray i respectively.
At the feed stage (NF=7) we have:

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 18:06:43 UTC from IEEE Xplore. Restrictions apply.
 d (3) composition changes due to the disturbances in feed. It has
M NF = L NF +1 − L NF + V NF −1 − V NF + F been shown that the composition response is the slowest
dt
dynamic of the column [7], and it is also one of the objectives
and in controlling the column.
d
(M NF x NF ) = L NF +1 x NF +1 − L NF x NF + V NF −1 y NF − The simulation was carried out in MATLAB. First, a steady
dt (4)
state was reached by simulating the column for 20000
− V NF y NF −1 + Fz F
minutes. The steady state data are shown in Table 1

in which F is the feed flow rate and zF is the concentration of TABLE I. STEADY STATE DATA OF THE COLUMN
the light component in the feed. zF α N NF xD xB D B
The reboiler is also an equilibrium stage with i=1: 0.5 1.5 15 7 0.8837 0.1163 0.5 0.5

d (5) After the column reached steady state, we carried out

M 1 = L2 − V1 − B
dt experiments to test the responses of the compositions under
d different disturbances. A step change in feed flow rate was
M 1 x1 = L2 x 2 − V1 y1 − Bx1 . (6)
chosen since this is the most frequent disturbance in reality.
dt The magnitudes of the step change were varied and the changes
At the condenser we have i = NT = 16 and in the compositions of the light component at the reboiler (xB)
d (7) and at the condenser (xD) were captured. The levels of distillate
M NT = V NT −1 − L NT − D
dt and bottom tanks must be kept constant during simulation by
d manipulating the distillate (D) and bottom (B) flows. In the
(M NT x NT ) = VNT −1 y NT −1 − L NT x NT − Dx NT (8) model these level controls were done by proportional control
dt with gain equal to 10.
where B is the bottom product flow rate and D is the distillate
product flow rate. Fig. 2 shows the changes in compositions due to 1%
The composition of the heavy component is related to the increases in feed flow rate.
composition of the light component via the relative volatility
formula:
αx i (9)
yi = .
1 + (α − 1)xi
The liquid flow dynamics is considered due to its important
effect on the initial response of the column. The formulas for
the liquid holdup are:
M − M 0i (10)
Li = L0b + i + (Vi −1 − V0 )λ
τ
for i from 2 to NF and
M − M 0i (11)
Li = L0 + i + (Vi −1 − V0t )λ
τ
for i from NF+1 to NT-1
where L0 is the nominal reflux flow and M0i is the nominal
reboiler holdup (kmol) on stage i. These values are achieved Figure 2. Composition responses to 1% step increase in feed flow rate
after we do steady state simulation (see Table 1). τ is the time
constant for liquid dynamics, in this model it is chosen to be For 1% increase in feed flow rate, the response of xD has a
0.063 (min), and λ represents the effect of vapor flow on liquid time constant of about 17 minutes while the response of xB has
flow. In the simulation we ignore this effect by setting λ=0. L0b a time constant of 11 minutes. The curves resemble the
is the nominal liquid flow below feed, given by the formula: response of a first order system with different time constants
(the time it takes for the signal to reach 63% of its final value),
L0b = L0 + q F 0 F0 (12) which is consistent with other works [7],[8].
in which F0 = 1(kmol/min) is the nominal feed rate, qF0 = 1 is Fig. 3 shows the results for 5% increase in feed flow rate,
the nominal fraction of liquid in the feed. which are 15 minutes and 10 minutes respectively. We can see
the responses in this case are faster than in the case of 1%
increase in feed flow rate.
IV. SIMULATION RESULTS AND DISCUSSION
The purpose of the simulation is to examine the dynamics Two simulations were done with 0.5% and 15% step
behavior of the column. In this paper we focus on the product changes. The results are shown in Table 2.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 18:06:43 UTC from IEEE Xplore. Restrictions apply.
 TABLE II. TIME CONSTANTS (IN MINUTES) OF THE COMPOSITION TABLE III. TIME CONSTANTS (IN MINUTES) OF THE COMPOSITION
CHANGES DUE TO DIFFERENT FEED FLOW RATE PERTURBATIONS CHANGES DUE TO DIFFERENT FEED COMPOSITION PERTURBATIONS

0.5% 1% 5% 15% 0.5% 1% 5% 15%

xD 18 17 15 14 xD 14 13 13 12
xB 10 11 10 9 xB 14 15 14 14

The time constants don’t vary much when we vary the

magnitudes of the disturbances. By comparing the steady-state
gains between the response due to change in feed flow rate and
the response due to change in feed composition, we can see the
effect of change in feed composition is greater than the effect
of change in feed flow rate.
Another important dynamics of the column is the effect of
internal and external flows on the compositions of the
products. We simulate a change in external flow by increasing
the reflux rate L while keeping the boilup rate V constant. The
composition response to 0.1% increases in L and V is shown
in Fig. 5.

Figure 3. Composition responses to 5% step increase in feed flow rate

From the simulation results we can observe the slow

respond of the composition of the distillation columns which
has been acknowledged by many practical and theoretical
works [8]. The results were compared with the short-cut
formula introduced by Skogestad and Morari [7]:
MI
τ sc = (13)
I s ln S
NT
where: M I = ∑M
i=2
i is the total holdup of liquid inside the

column, Figure 4. Composition response to 1% increase in feed composition

I S = Dx D (1 − x D ) + Bx B (1 − x B ) is the “impurity sum”
x D (1 − x B )
and S = is the separation factor.
(1 − x D )x B
Substituting the steady state values from Table 1 to (13)
above we get τ = 18 (minutes)
This time constant is consistent with the ones obtained from
simulation of 0.5% and 1% increases in feed flow, but it is not
consistent with the cases when the feed flow increases are 5%
and 15%. From here we can conclude that (13) is valid only
for small perturbations.
In the next step, we test the composition responses to
disturbances in the feed composition. Step increases of 0.5%,
1%, 5% and 15% were applied to the composition of the light
component in the feed. The composition responses in this case
also resemble a first-order response. The response to 1% step
increase in feed composition is plotted in Fig. 4.
Table 3 shows the time constants of the responses. Figure 5. Composition changes due to 0.1% increases in L and V

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 18:06:43 UTC from IEEE Xplore. Restrictions apply.
 In the case of increasing L with V being constant we can see Nomenclature
the changes in liquid compositions of the light component in B bottom product flow rate (kmol/min)
the reboiler (xB) and in the condenser (xD) are both positive. D distillate product flow rate (kmol/min)
This means the light component gets purer in the distillate tank F feed flow rate (kmol/min)
and the heavy component gets more impure in the bottom tank. IS impurity sum
In the opposite case, when we increase V and keep L constant
we get the opposite result: the heavy component gets purer in L liquid flow rate (kmol/min)
the bottom tank and the light component gets more impure in M liquid holdup (kmol)
the distillate tank. In both cases the time constants are close to MI total liquid holdup inside the column (kmol)
14 minutes. N number of physical trays (N=15)
NF tray number at feed position (NF=7)
An internal flow was simulated by applying a simultaneous
increase in L and V, while keeping the product rates, D and B, NT number of theoretical trays, including the
constant. Fig. 6 shows the composition response to 10% condenser
increase in both L and V. S separation factor
V vapor flow rate (kmol/min)
x liquid composition of the light component
(mol fraction)
y vapor composition of the light component
(mol fraction)

Greek letter
α relative volatility
λ parameter for the effect of vapor flow to
liquid flow
τ time constant for liquid flow dynamics

Subscripts
B Bottom tank
D Distillate tank
i tray number
Figure 6. Composition changes due to 10% increase in both L and V S sum
From the graph we can see xB is negative while yD is sc short cut
positive, meaning that both products get purer in this case. We
can also observe the difference between the effects of internal REFERENCES
flow and external flow. A small change in external flow (0.1%)
will make the products compositions change with different [1] Z. Abdullah, N. Aziz and Z. Ahmad.(2007) “Nonlinear Modelling
directions while a much bigger change in internal flow (10%) Application in Distillation Column”. Chemical Product and Process
will make the products compositions change with the same Modelling: Vol. 2:Iss. 3, Artical 12. Available at:
direction. This is consistent with the results obtained by [9]. http://www.bepress.com/cppm/vol2/iss3/12
[2] S. Skogestad. “Dynamics and Control of Distillation Columns – A
Critical Review”. Presented at IFAC-symposium DYCORD+’92,
V. CONCLUSION AND FUTURE DIRECTIONS Maryland, Apr. 27-29, 1992
[3] W. L. Luyben. “Derivation of Transfer Functions for Highly Nonlinear
In this paper a model of a continuous binary distillation was Distillation Columns”. Ind. Eng. Chem. Res. 1987, 26, 2490-2495
constructed based on mass balance and constant relative [4] S. Skogestad. “Modelling and Dynamic Simulation for Process Control:
volatility. The model was then used to simulate the dynamics Issues in dynamic modeling using distillation as an example”. Lecture
of the column. The simulation results show that the notes at seminar on “Modelling and Optimization of Chemical
composition responses are close to the response of a first order Processes” NTH, Trondheim, 26-28 August 1991.
[5] B. Wittgens and S. Skogestad. “Evaluation of Dynamic Models of
system. The response to a change in feed flow rate can be Distillation Columns with Emphasis on the Initial Response”.
formulated if the change is small. The response due to change Modelling, Identification and Control, 2000, vol. 21, no. 2, 83-103
in feed composition has larger gain than the response due to [6] E. W. Jacobsen. “Studies on Dynamics and Control of Distillation
change in feed flow rate. Columns”. Ph.D. dissertation. Uni. of Trondheim, Trondheim, Norway.
Dec. 1991.
The next activities will be modeling and simulations of [7] S. Skogestad and M. Morari. “The Dominant Time Constant for
more rigorous models, and observing the responses of the Distillation Columns”. Comput. Chem.. Engng, Vol. 11, No. 6, pp. 607-
compositions to other disturbances. Experiments on the real 617, 1987.
[8] C. Fuentes and W. L. Luyben. “Control of high-purity Distillation
plant will also be carried out in order to validate the model. Columns”. Ind. Eng. Chem. Process. Des. Dev. 361, 1983
Future work also includes developing a suitable multivariable [9] S. Skogestad. “Dynamics and Control of Distillation Columns: A
control structure (such as a 2x2 control structure) to the Tutorial Introduction”. Trans. IChemE, Vol. 75, Part A, Sept. 1997
distillation column.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 18:06:43 UTC from IEEE Xplore. Restrictions apply.