Modeling and Control of Distillation Column in a
Petroleum Process
Vu Trieu Minh
Mechanical Engineering Department
Universiti Teknologi PETRONAS (UTP)
31750 Tronoh, Perak Darul Ridzuan, MALAYSIA
vutrieuminh@petronas.com.my
Abstract — This paper introduces a calculation procedure for
modeling and control simulation of a condensate distillation
column based on the energy balance (L-V) structure. In this
control, the reflux rate L and the boilup rate V are used as the
inputs to control the outputs of the purity of the distillate
overhead and the impurity of the bottom products. The
mathematical modeling simulation is an important part for the
process dynamic analysis and the plant initial design. In this
paper, the mathematical modeling simulation is accomplished
over three phases: the basic nonlinear model of the plant; the
full-order linearised model; and the reduced-order linear model.
The reduced-order linear model is then used as the reference
model for a model-reference adaptive control (MRAC) system to
verify the applicable ability of a conventional adaptive controller
for a distillation column dealing with the disturbance and the
model-plant mismatch as the influence of the plant feed
disturbances.
Keywords - Distillate purity/impurity; Concentration control;
Basic nonlinear model; Full-order linearised model; Reduced-order
linear model; Model-reference adaptive control.
I. INTRODUCTION
Distillation is the most popular and important separation
method in the petroleum industries for purification of final
products. Distillation columns are made up of several
components, each of which is used either to transfer heat
energy or enhance mass transfer.
Calculation of the distillation column in this paper is based
on a real petroleum project to build a gas processing plant to
raise the utility value of condensate. The quality of the output
products is the purity of the distillate, xD, higher/equal than
98% and the impurity of the bottoms, xB, less/equal than 2%.
The basic feed stock data and its actual compositions are based
on the reference [1].
The L-V structure, which is called energy balance
structure, can be considered as the standard control structure
for a dual composition control distillation. In this control
structure the liquid flow rate L and the vapor flow rate V are
the control inputs. The objective of the controller is to
maintain the product outputs concentrations xB and xD despite
the disturbance in the feed flow F and the feed concentration
cF (Figure 1).
This work was supported by Universiti Teknologi PETRONAS (UTP)
978-1-4244-5046-6/10/$26.00 2010
c IEEE
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 14:15:06 UTC from IEEE Xplore. Restrictions apply.
The goals of this paper are twofold: first, to present a
theoretical calculation procedure of a condensate column for
simulation and analysis as an initial step of a project feasibility
study, and second, for the controller design: a reduced-order
linear model is derived such that it best reflects the dynamics
of the distillation process and used as the reference model for
a model-reference adaptive control (MRAC) system to verify
the ability of a conventional adaptive controller for a
distillation process dealing with the disturbance and the plant-
model mismatch as the influence of the feed disturbances.
In this study, the system identification is not employed
since experiments requiring a real distillation column is still not
implemented yet. All calculations and simulations in this paper
are implemented by Matlab version 7.0.
Condenser
Reflux Drum
Input u1 Distillate Flow D
Reflux Rate L Output xD
Feed Flow F
Boilup Rate V
Input u2
Bottom Flow B
Reboiler Output xB
Figure 1. Distillation Flowsheet
II. MATHEMATICAL PROCESS MODEL AND SIMULATION
The column is designed with N=14 trays. The model is
simplified by lumping some components together
(pseudocomponents) and modeling the column dynamics on
these pseudocomponents only [2].
Table 1 summaries the initial calculated data for the main
streams of input feed flow rate: Condensate, output distillate
overhead product: LPG and output bottom product: Raw
gasoline.
The vapor boilup V generated by the heat input to the
reboiler is calculated as [3]. The latent heat at any temperature
is described in terms of the latent heat at the normal boiling
point [4]. Major design parameters to determine the liquid
259
 holdup on tray, column base and reflux drum are calculated M B x1 = ( L + LF ) x2 − Vy1 − Bx1 (7)
mainly based on references [5-7]. Although the model is simplified, the representation of the
TABLE 1. Main Streams distillation system is still nonlinear due to the vapor-liquid
Stream Condensate LPG Raw gasoline
Temperature (0C) 118 46 144
equilibrium relationship between yn and xn in (1).
Pressure (atm) 4.6 4.0 4.6 The distillation process simulation is shown in Figure 2.
Density (kg/m3) 670 585 727
Volume flow rate (m3/h) 22.76 8.78 21.88 If there is no disturbance in the operating conditions as
Mass flow rate (kg/h) 15480 5061 10405 shown in Figure 3, the system is to reach the steady state such
Plant capacity (ton/year) 130000 43000 87000
that the purity of the distillate product xD = 0.9654 and the
The model is simplified under assumptions in [8].
Constant relative volatility throughout the column and the impurity of the bottoms product xB = 0.0375 .
vapor-liquid equilibrium relation can be expressed as:
Table 2 indicates the steady state values of concentration of
α xn xn and yn on each tray.
yn = (1)
1 + (α − 1) xn
th
where xn : liquid concentration on n stage; yn : vapor 1
xD
concentration on nth stage; α : relative volatility. 0.9

0.8
In1 Out1

Purity of the Distillate Product x

Condenser & Reflux Drum
In1 Out1
0.7
In2 Out2
Tray 14
In1 Out1 0.6
In2 Out2 LPG Purity
Tray 13
In1 Out1 0
In2 Out2 0.5
OUTPUT 1
Tray 12
MODULE OF RECTIFYING SECTION In1 Out1
In2 Out2
Tray 11
0.4
In1 Out1
In2 Out2
Tray 10 0.3
In1 Out1
In2 Out2
Tray 9
0.2
In1 Out1
In2
11.3343 In3 Out2
Tray 8 0.1
yF*VF
FEED RATE xB
0
IN In1 0 50 100 150 200 250 300 350
Out1
4.6903
In2
In3 Out2
Time
Tray 7
xF*LF
In1 Out1

Figure 3. Steady State Values of Concentrations xn on each Tray

In2 Out2
Tray 6
In1 Out1
In2 Out2

MODULE OF STRIPPING SECTION In1

Tray 5
Out1 Since the feed stream depends on the upstream processes.
The changes of the feed stream can be considered as
In2 Out2
Tray 4
In1 Out1
In2

In1
Tray 3
Out2

Out1
Gasoline Impurity
0
disturbances including the changing in feed flow rates and
In2
Tray 2
Out2
OUTPUT 2 feed compositions.
In1 Out1
In2 Out2

In1
Tray 1
Out1 TABLE 2. Steady State Values of Concentrations xn and yn on each Tray
Out2
Column base & Reboiler

Stage Bottom Tray 1 Tray 2 Tray 3 Tray 4 Tray 5 Tray 6 Tray 7

xn 0.0375 0.0920 0.1559 0.2120 0.2461 0.2628 0.2701 0.2731
yn 0.1812 0.3653 0.5120 0.6044 0.6496 0.6694 0.6776 0.6809
Figure 2. Model Simulation with Matlab Simulink Tray Tray Tray Tray Tray
Stage Tray 8 Tray 9 Distillate
10 11 12 13 14
The dynamic model can be expressed by the following xn 0.2811 0.3177 0.3963 0.5336 0.7041 0.8449 0.9369 0.9654
equations: yn 0.6895 0.7256 0.7885 0.8666 0.9311 0.9687 0.9883 0.9937
Condenser (n=N+2): Simulations with these disturbances indicate that the
M D xn = (V + VF ) yn −1 − Lxn − Dxn (2) quality of the output products gets worse if the disturbances
exceed some certain ranges as shown in Table 3.
Tray n (n=f+2 to N+1):
Mxn = (V + VF )( yn −1 − yn ) + L( xn +1 − xn ) (3) TABLE 3. Product Quality Depending on the Change of the Feed Rates
Tray above the feed flow (n=f+1): Purity of the Distillate Impurity of the Bottoms
Mxn = V ( yn −1 − yn ) + L( xn +1 − xn ) + VF ( yF − yn ) (4) Product xD (%) Product xB (%)
Tray below the feed flow (n=f): Normal Feed Rate 96.54 3.75
Mxn = V ( yn −1 − yn ) + L( xn +1 − xn ) + LF ( xF − xn ) (5) Reduced Feed Rate 10% 90.23 0.66
Increased Feed Rate 10% 97.30 11.66
Tray n (n=2 to f-1):
The designed system does not achieve the operational
Mxn = V ( yn −1 − yn ) + ( L + LF )( xn +1 − xn ) (6) objective of the product quality ( xD ≥ 0.98 and xB ≤ 0.02 ) and
Reboiler (n=1): the product quality will get worse dealing with disturbances.

260 2010 5th IEEE Conference on Industrial Electronics and Applicationsis

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 14:15:06 UTC from IEEE Xplore. Restrictions apply.
 Hence we will use an adaptive controller - MRAC to take the IV. MRAC BUILDING AND SIMULATION
system from these steady state outputs to the desired output Adaptive control is widely applied in petroleum industries
targets ( xD ≥ 98% and xB ≤ 2% ). because of two main reasons: Firstly, most of processes are
nonlinear and the linearized models are used to design the
III. LINEARIZATION OF THE DISTILLATION PROCESS
controllers, so that the controller must change and adapt to the
In order to obtain a linear control model for this nonlinear model-plant mismatch; Secondly, most of the processes are
system, we assume that the variables deviate only slightly from non-stationary or their characteristics are changed with time,
some operating conditions [9]. Then the nonlinear equation in this leads again to adapt the changing control parameters.
(1) can be expanded into a Taylor’s series. If the variation The general form of a MRAC is based on an inner-loop
xn − xn is small, we can neglect the higher-order terms in Linear Model Reference Controller (LMRC) and an outer
xn − xn The linearization of the distillation column leads to a adaptive loop shown in Figure 4. In order to eliminate errors
16th order linear model in the state space form: between the model and the plant and the controller is
asymptotically stable, MRAC will calculate online the
z (t ) = Am z (t ) + Bm u (t ) adjustment parameters in gains L and M by θ L (t ) and
, where:
y (t ) = Cm z (t ) θ M (t ) as detected state error e(t ) when changing A , B in the
ª x1 (t ) − x1 Steady State º process plant.
« x (t ) − x » Simulation program is constructed using Maltab Simulink
ª L(t ) − LSteady State º
z(t ) = «
2 2 Steady State »
, u(t ) = « », (8) with the following data:
¬«V (t ) −VSteady State ¼»
« # »
« » Process Plant:
«¬ x16 (t ) − x16 Steady State »¼ z = Az + Bu + noise ªα1 0 º ª β1 0 º
where A = « » , B=« »,
ª x1 (t ) − x1 Steady State º y = Cz ¬ 0 α2 ¼ ¬ 0 β2 ¼
y(t ) = « »
¬ x16 (t ) − x16 Steady State ¼ ª 0.004 −0.007 º
C=« » and α1 , α 2 , β1 , β 2 are changing
The full-order linear model which represents a two ¬ −0.0011 0.0017 ¼
inputs – two outputs plant in equation in (8) can be expressed and dependent on the process dynamics.
as a reduced order linear model as in [10-11]:
ª xD º 1 ªLº
«x » = G (0) « » (9)
¬ B ¼ 1+τ cs ¬V ¼
where: G(0) is the steady state gain: G (0) = −CA−1 B , τ c is
the time constant:
MI M D (1 − xD ) xD M B (1 − xB ) xB
τc = + + where MI
I s ln S Is Is
(kmole): the total holdup of liquid inside the column; M D
(kmole): liquid holdup in the condenser; M B (kmole): liquid
holdup in the reboiler; I s : the “impurity sum”; and S : the
separation factor.
As the result of calculation, the reduced-order linear
model of the plant is a first order system with a time constant
of τ c = 1.9588 (h) :
ª xD º 1 ª 0.0042 −0.0062º ª L º
«x » = (10) Figure 4. MRAC Block Diagram
« »« »
¬ B ¼ 1 + 1.9588s ¬ −0.0052 0.0072 ¼ ¬V ¼ Reference Model:
Equation (10) is equivalent to the following linear model
zm = Am zm + Bm uc ª −0.2616 0 º
in state space: where Am = « » ,
−0.5105 0 1 0 y m = Cm z m ¬ 0 −0.2616 ¼
zr (t ) = zr (t ) + u (t )
0 −0.5105 0 1 ª1 0 º ª 0.004 −0.007 º
(11) Bm = « » , Cm = « −0.0011 0.0017 »
0.0021 −0.0031 ¬0 1 ¼ ¬ ¼
yr (t ) = zr (t ) State Feedback:
−0.0026 0.0037
ªθ 0 º ªθ3 0 º
Stability test: The system is asymptotically stable since all u = Muc − Lz where L = « 1 » and M = « ».
eigenvalues of the state matrix are in the left half of the ¬ 0 θ2 ¼ ¬ 0 θ4 ¼
complex plane ( [ −0.5105, − 0.5105] ). Closed Loop:
z = ( A − BL) z + BMuc = Ac (θ ) z + Bc (θ )uc

2010 5th IEEE Conference on Industrial Electronics and Applicationsis 261

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 14:15:06 UTC from IEEE Xplore. Restrictions apply.
 Error Equation: steady-state outputs to which the regulator converges, that
ªe º minimize their deviation from the desired target values. Target
e = z − zm = « 1 » is a vector of state errors, tracking can be formulated as an optimization MPC problem
¬ e2 ¼
that uses the quadratic objective function to minimize the
Lyapunov Function:
deviation of the steady-state outputs and inputs from the
1
V (e,θ ) = ( γ eT Pe + (θ − θ 0 )T (θ − θ 0 ) ) where γ is an desired target values:
2
°­
N y −1
ª( yt + k |t − r )' Ξ( yt + k |t − r )º °½
adaptive gain and P is a chosen positive matrix. min ® J (U , x(t )) = ¦ « »¾
'
Derivative Calculation of Lyapunov Fuction: U {u1 ,...,ut +Nu −1 }
°¯ k =0 ¬«+Δut +k ΨΔut + k ¼» ¿°
dV γ § dθ · subject to:
= − eT Qe + (θ − θ 0 )T ¨ + γΨT Pe ¸ where
dt 2 © dt ¹ yt+k|t ∈[ymin, ymax] for k =0,1,…Ny -1;
dV ut+k|t ∈[umin, umax] and ǻut+k|t ∈[ǻumin, ǻumax]
Q = − AmT P − PAm and for the stability of the system, < 0. for k =0,1,…Nu -1, (14)
dt
Parameters Adjustment: ut + k = 0 and Δut + k = 0 for k ≥ N u
ª − β1 z1 0 º ª dθ1 / dt º ª γβ1 z1e1 º with N u , N y are the control and prediction horizons
« 0 »
− β 2 z2 » « dθ / dt » « 2γβ z e » and the update space state:
dθ ªe º
= −γ «
« β c1u1 0 »
[ P] « 1 » = « 2
e d θ / dt
»=«
−γβ
2 2 2 »
xt |t = x(t ), xt + k +1|t = Axt + k |t + But + k , yt + k |t = Cxt + k |t
dt 1uc1e1
¬ 2¼ « 3
» « »
« » « » « » where yt + k |t and r are the predicted outputs and the output
¬ 0 β 2 u2 c ¼ ¬ dθ 4 / dt ¼ ¬ −2γβ 2 uc 2 e2 ¼
Simulation results and analysis: setpoints, respectively. Ξ and Ψ are penalty matrices. In the
We assume that the reduced-order linear model in equation target tracking MPC regulator, the steady-state outputs of the
(11) can also maintain the similar steady state outputs as the process will be equal to the target setpoints if there is no
basic nonlinear model. Now we use this model as an MRAC to constraint and disturbance. The formulation (14) is the one the
authors consider for the remainder of this paper to verify the
take the process plant from these steady state outputs to the
ability of a MPC for this process.
desired targets amid the disturbances and the plant-model
mismatches as the influence of the feed stock disturbances.
MRAC
The design of a new adaptive controller is shown in Figure
5 where we install an MRAC and a closed-loop MPC (Model
Predictive Control) controller to eliminate the errors between Setpoint + Output
the reference setpoints and the outputs. MPC is also known as MPC Process
_
Receding Horizon Control or Moving Horizon Control. It is an
optimization-based technique to generate online feedback
optimal control to linear or nonlinear systems subject to Noise
constraints. The feedback control is obtained by solving online
a sequence of open-loop optimal control problems.
Calculations for the MPC controller in this paper are referred
to several physical restrictions, especially in petroleum Figure 5. Adaptive Controller with MRAC and MPC
industry [13]. The constraints on the inputs and outputs can be
We run this controller system with different plant-model
grouped together into a single matrix inequality as:
ª −0.50 0 º
Δumin ≤ Δu (k + i ) ≤ Δumax , i = 0, ... , m − 1 mismatches, for instance, a plant with A = « ,
¬ 0 −0.75»¼
umin ≤ u (k + i ) ≤ umax , i = 0, ... , m − 1 (12)
ª1.5 0 º
ymin ≤ yˆ(k + i / k ) ≤ ymax , i = 1, ... , p B=« » and an adaptive gain γ = 25 . The operating
¬ 0 2.5¼
It is important that the controller satisfies these constraints. setpoints for the real outputs are xDR = 0.99 and xBR = 0.01 .
Through some standard matrix algebra, these constraints can For the MPC controller, the sampling time has been chosen as
be defined together in the form of a single inequality. Then T  1.0 second, the final values chosen as the prediction
the constraint set in equation (12) can be represented in terms horizon N y = N u = 5 , the penalty matrices are chosen as
of the following inequality:
ΠΔu + ϒ ≥ 0 (13) ª2 0º ª1 0 º
Ξ=« » and Ψ = « » . Other parameters are referred
Processes usually operate with the desired target values or ¬0 2¼ ¬0 1 ¼
the output setpoints. In some operations, the desired setpoint to as in [14]. Simulation in Figure 6 shows that the controlled
values might change over time. Therefore, the target tracking outputs xD and xB are always stable and tracking to the
is an important part of any controller design. The objective model outputs and the reference setpoints (the dotted lines)
function for the target tracking is to determine the feasible amid the disturbances and the plant-model mismatches.

262 2010 5th IEEE Conference on Industrial Electronics and Applicationsis

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 14:15:06 UTC from IEEE Xplore. Restrictions apply.

0.03
rD
0.02
Purity of the Distillate Product x
0.01
0
-0.01
-0.02
r x
B [2] Kehlen, H., and Ratzsch, M. T., “Complex Multicomponent Distillation
-0.03
20 40
Figure 6. Correlation of Plant, Model Outputs and Reference Setpoints
V.
The mathematical modeling simulation is accomplished
over three phases: the basic nonlinear model, the full order
linearized model and the reduced order linear model. Results
from the simulations and analysis are helpful for initial steps
of a petroleum project feasibility study and design.
The reduced order linear model is used as the reference
model for an MRAC controller. The controller of MRAC and
MPC theoretically allows the plant outputs tracking the
reference setpoints to achieve the desired product quality amid
the disturbances and the model-plant mismatches as the
influence of the feed stock disturbances.
In this paper, the calculation of the mathematical model
building and the reduced-order linear adaptive controller is
only based on the physical laws from the process. The real
system identifications including the experimental production
factors, specific designed structures, parameters estimation and
the system validation are not mentioned here. Further, the
MRAC controller is not suitable for the on-line handling of the
process constraints.
2010 5th IEEE Conference on Industrial Electronics and Applicationsis
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DE SANTA CATARINA. Downloaded on April 08,2021 at 14:15:06 UTC from IEEE Xplore. Restrictions apply.
ACKNOWLEDGMENT
xD
The authors would like to thank the comments provided by
the anonymous reviewers and editor, which help the authors
improve this paper significantly. The authors have taken into
consideration all comments of the reviewers in the final
version of the paper. This work was supported by Universiti
Teknologi PETRONAS (UTP).
REFERENCES
[1] PetroVietnam Gas Company, Condensate Processing Plant Project –
Process Description Document No. 82036-02BM-01. PetroVietnam, pp. 1-
54 (1999).
B
60 80 100 120 140
Calculations by Continuous Thermodynamics”. Chem. Eng. Sci., pp. 221-
Time 232 (1987).
[3] Franks, R. G.E., Modeling and Simulation in Chemical Engineering.
Wiley-Interscience, N.Y., (1972).
CONCLUSION [4] Nelson, W. L., Petroleum Refinery Engineering. Auckland McGraw-Hill,
(1982).
[5] Joshi, M. V., Process Equipment Design. New Delhi, Macmillan Company
of India, (1979).
[6] McCabe, W. L. , and Smith J. C., Unit Operations of Chemical
Engineering. N.Y McGraw-Hill, (1976).
[7] Wuithier, P., Le Petrole Raffinage et Genie Chimique. Paris Publications
de l’Institut Francaise du Petrole, (1972).
[8] Stephanopoulos, G., Chemical Process Control. Prentice Hall
International, (1984).
[9] Katsuhiko Ogata, Model Control Engineering. Prentice-Hall International
(1982).
[10] Papadouratis, A., Doherty, M. F., and Douglas, J. M. “Approximate
Dynamic Models for Chemical Process Systems”. Ind. Eng. Chem. Res.,
pp. 546-522 (1989).
[11] Skogestad, S., and Morari, M. “The Dominant Time Constant for
Distillation Columns”. Comp. Chem. Eng., pp. 607-617 (1987).
[12] Marie, E et al., “Coordinator MPC for Maximizing Plant Throughput”.
Comp. Chem. Eng., pp. 195-204 (2008).
[13] Vu Trieu Minh and Nitin Afzulpurkar. ‘Robustness of Model Predictive
Control for Ill-conditioned Distillation Process’. International Journal of
Developments in Chemical Engineering and Mineral Processing, Vol. 13,
No.3/4. pp. 151-158. (2005).
[14] Vu Trieu Minh and Ahmad Majdi Abdul Rani. ‘Modeling and Control
Simulation for a Petroleum Process’. Mathematical Problems in
Engineering. Vol. 2009. Article ID MPE/404702, 16 pages, (2009).
263