Modelling and Dynamics of an Extractive Distillation

Column
S . KUMAR, J . D. WRIGHT‘ and P. A . TAYLOR

McMaster University, Department of Chemical Engineering, Hamilton, Ontario, Canada L8S 4L7
Much of the published research literature on distillation column dynamics and control has been focussed on binary
systems or on multicomponent systems in which key components are used to define the operations. This paper presents
steady state and dynamic non-linear mass and energy balance models and corresponding experimental results for an
extractive distillation system separating an acetone-methanol azeotrope using a water solvent. Murphree efficiencies are
used to corroborateexperimental data. The system comprises a 15 cm diameter bubblecap column with I3 trays. Transfer
function models fitted to process data are compared with mass and energy balance simulation models describing the
transient responses of the system at various operating conditions to changes in the inlet feedrates, feed composition,
reflux rate and reboiler duty. The nonlinear behaviour of the column is illustrated by gain changes at different operating
conditions, reinforcing the need for comprehensive models for complete system dynamic characterization. The system
is monitored and controlled by a V A X I1/750 process computer with a DPM-23 distributed front-end using on-line
temperature and composition measurements.

La majeure partie de la litttrature qui a paru sur la dynamique et le contrde des colonnes de distillation a etC concentree
sur des systemes binaires ou multicomponents dans lesquels on emploie les principaux composants pour definir les
opkrations. On souligne, dans le present travail, des modeles d’equilibre de masse et d’energie stables et dynamiques,
non lineaires, ainsi que les resultats experimentaux correspondants pour un systeme de distillation skparant un melange
azeotrope d’acktone-methanol,dans lequel on emploie un solvant aqueux. On emploie les efficacitts de Murphree pour
corroborer les rksultats expkrimentaux; le systeme comprend une colonne a double-calotte d’un diametre de 15 centi-
metres et impliquant 13 plateaux. On compare les modeles qui s’adaptent aux donnees de traitement avec les modeles
de simulation d’equilibrede masse et d’knergie ou I’on dCcrit les rkponses transitoires du systeme a differentes conditions
op6ratoires au changements dans les vitesses d’alimentation aux entrees, a la composition d’alirnentation, au taux de
reflux et au service du brouilleur. On illustre le comportement non lineairc de la colonne par des changements dans les
gains a differentes conditions opkratoires, ce qui augmente le besoin de modeles approfondis pour la caracterisation
dynamique du systeme complet. Le systeme est surveil16 et contrble par une calculatrice de traitement (process
computer) V A X I1/750,impliquant un DPM-23 repartie a I’extrhe-avant, en employant un calcul integre de mesures
de temp6ratures et de compositions.

E xtractive and azeotropic distillation are important unit

operations for separation of either close-boiling or azeo-
tropic mixtures in the petrochemical industries. Common
ified calculation procedure is only valid when individual
stage efficiencies are unity and the k-values are not strongly
dependent upon the vapour phase compositions. They used
examples include recovery of butadiene from butanes, a Newton-Raphson iteration procedure to solve the equa-
butenes, butynes and 1,2-butadiene, and separation of pro- tions. Hirose et al. (1980) discussed an algorithm which is
pylene from propane by extractive distillation. Introduction valid when the reflux ratio is not specified but the reboiler
of a third component is used either to increase the relative duty or the condenser duty is given. However, this method
volatilities of the key components or to break azeotropes requires more computer time than Christiansen et al. (1979).
which could otherwise not be separated. Selection of a third The work of Burman and Maddox ( 1969), Osborne ( 1965),
component for these mixtures is discussed by Berg (1969). and by Tetlow et al. (1966) formed the basis for much of the
Applications of extractive distillation in industry have been subsequent work on the dynamics of multicomponent sys-
discussed by Buck et al. (1969), and Gerester (1969) has a tems. Marr (1962) used parameters, which were a con-
short comparison of the two techniques. Most of the pub- tinuous function of the vertical column position, to incorpo-
lished results are based on steady state simulations, but for rate factors from the geometrical and mechanical design of
control purposes, it is necessary to investigate the dynamic the system. This resulted in a rather complex model, but it
behaviour. did include the dynamic effects. Ballard et al. (1978)
Many dynamic and control studies have been done on the presented an algorithm which required calculation of the
distillation of binary mixtures. See for example the survey Jacobian matrix at the beginning of each time step. This was
papers by Waller (1982) and a representative paper such as used to implement a second-order Runge Kutta algorithm
Tung and Edgar (1979). The modelling of multicomponent for solving the dynamic equations.
distillation systems has not received nearly as much atten- Most of the dynamic studies use either a tray-to-tray con-
tion, particularly in the dynamic sense. Christiansen et al. cept to model the column or a grouping of trays as a column
(1979) used a modified Naphtali Sandoholm Column Calcu- section. The latter form is used in the application of bilinear
lation Procedure coupled with the Soave-Redlich Kwong models for a three-component system studied by Benallow
Equation of State for generating k-values and enthalpies to et al. (1982). An alternative form has been used by Stewart
model a multicomponent distillation. However, this mod- et al. (1981) where orthogonal collocation techniques have
been used to transform the difference-differential equations
‘Xerox Research Centre of Canada, 2660 Speakman Drive, Mis- into a set of partial differential equations which were then
sissauga, Ontario, Canada L5K 2L1. solve numerically. These techniques have also been applied

780 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 62. DECEMBER 1984
by Wong and Luus ( 1980) and by Cho and Joseph (1980).
None have been compared with experimental data and none
have been used for extractive distillation systems. In general
terms, the use of sections of the column gives a relatively
low-order model which is very useful for on-line control
purposes. A tray-to-tray method, on the other hand, gives
more physical insight as to what is really happening in the
column.
Since the long range objectives of this work are to devel-
op models for complex distillation columns suitable for con-
trol, it is useful to examine briefly the control literature for
distillation systems. Most of the published work in dis-
tillation control has been based on simulation studies of
column dynamics. Early work by Rosenbrock (1962) has
been followed by a number of studies by Davison (see for
example Davison (1970)). By contrast, Wood and Berry
(1973) discussed transfer function models derived from pro-
cess data for a binary system. They did not compare results
to mass and energy balance models although several control
1
strategies based on the transfer function models were de-
rived. Tung and Edgar (1979) presented experimental data
along with model simulations written in multivariable state SG
space format, again only for a binary system. Patke and
Deshpande (1982) presented a comparative assessment of
inferential and parallel cascade schemes for controlling top
product composition of a simulated multicomponent dis-
tillation column. They concluded that inferential control is
useful when only one composition is controlled. Luyben
(1975) showed that dual composition control results in min-
imum energy usage. McAvoy and Tsogas (1981) reported Figure 1 - Schematic diagram of the extractive distillation col-
that material balance control is superior to decoupling for umn.
dual composition control in high purity towers whereas
the reverse is true for the low purity columns. Waller and
Gustafsson (1972) reported that, in multicomponent sys- bubblecap. The tray spacing was 15 cm. A schematic dia-
tems, the non-key components can show most irregular gram of the column is shown in Figure l . The maximum
behaviour. All of this last group of papers is based on pressure rating for the column was 206.7 kPa but all experi-
simulation studies alone. Edgar and Schwanke (1976) note ments were carried out at atmospheric pressure. Tem-
that, even for binary distillation systems, the large dimen- peratures were measured for both vapour and liquid on each
sionality, incomplete measurement, unmeasured distur- tray in the column. Feed entered the column on the 10th tray
bances, and non-linearity make it difficult to implement (from the top) and the solvent stream entered on the 5th tray.
multivariable or adaptive control. One other difficulty is that A thermosyphon reboiler was used. Overhead product was
many controller design methods require precise models completely condensed, split and then reheated to provide for
which are very difficult to obtain for multicomponent sys- the reflux liquid stream to the column.
tems (especially for extractive columns where the system is The feed and solvent flowrates were maintained by con-
also multifeed). trol valves CVI and CV2. Two separate feed tanks were
Thus, the design of efficient and stable controllers for connected through solenoid vales, V I and V2 which gave
multicomponent distillation systems requires a simple the ability to introduce changes in feed composition directly
model which adequately describes the dynamic behaviour. by the computer system.
Unfortunately, experimental data on the dynamic behaviour The reflux flowrate and overhead product flowrate also
of extractive distillation columns are not available in the had individual control valves. Various combinations of con-
open literature. This has motivated the present work, which trol schemes for reflux ratio, condenser level control, and
compares both steady state and dynamic models, with the product flowrate could be configured through the on-line
corresponding experimental data for an extractive dis- computer system.
tillation system. It also addresses the nonlinearity issue. Temperatures for the feed and solvent lines were con-
trolled by SCR time proportioning controllers supplied by
Experimental system Barber Coleman. The controllers drove electrical resistance
heaters on the feedlines. The reflux temperature was also
In this study a binary azeotrope of acetone and methanol controlled using a time-proportioning SCR controller. How-
was broken using a water solvent. Selection of the solvent ever, since the objective was to maintain the reflux at the top
was made by the Scheibel method. Water was selected tray temperature, a thermocouple from the top tray was used
for reasons of cost, absence of toxicity problems, and to provide the setpoint for the feedback controller on the
functionality. reflux line. Both controllers were implemented in the com-
The experimental distillation system was comprised of a puter system.
15 cm diameter column fabricated from Quick Fit Glass A VAX 11/750 was used as a host computer to control
segments. There were 13 trays each with a single Teflon the column through a DPM 23 distributed front-end micro-

THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 62, DECEMBER 1984 78I
 The component material balance for the condenser is given
by
+
v2y12 = LlX,l DX,D . . . . . . . . . . . . . . . . . . . . . . . . (14
and for the reboiler is given by
V,tly,,Jtl= LJxiJ- BxIB . . . . . . . . . . . . . . . . . . . . . (le)
The enthalpy balance for the condenser is given by
V2H2 = Llhl + DHD + Qc .................. (If)
and for the reboiler is given by
VJ+lHltl= LJh, - Bhs + QR . . . . . . . . . . . . . . . . . (Ig)
Liquid and vapour specific enthalpies for the j t h tray are
defined by
N

Figure 2 - Typical tray configuration and nomenclature. HI = H,*y, . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2a)

)=I

hl = C ......................
h,,*xll - hmlxlng (2b)
I=I
computer system. The latter used a PDP 1 I /23 process 1/0
subsystem. An asynchronous serial communication line The performance of each tray is defined by the Murphree
linked a Varian Vista 401 Chromatographic Data System efficiency for each component
which provided analytical data from a process chro-
matograph for the liquid compositions. Separately, an Intel (3)
8085 Microcomputer System provided thermocouple scans
for about 30 temperatures in the distillation column. The equilibrium at the j t h tray for the ith component is
For the open loop step response experiments and for the defined by
steady state conditions, it was necessary to maintain the
reboiler and reflux condenser levels under closed loop con- y,* = yf:klJxlJ= K,x, . . . . . . . . . . . . . . . . . . . . . . . . (4)
trol. PI controllers were used to manipulate the bottoms where the activity coefficients, y I j Land the K-values, K,, are
product and overhead product flowrates. functions of composition and temperature, respectively.
Complete column details as well as a number of experi- Equations (3) and (4) may be combined to give
mental runs for this column have been described in detail by
Langille (1983). + Y I J + l ( l - E,) . . . . . . . . . . . . . . . . . (5)
yIJ= EIJKIJXll

Steady state simulation or

xij = (-yij+1[1 - El,] + y,)/(E,KiI) . . . . . . . . . . . . (6)
In a continuous distillation column, it is necessary to
write mass and enthalpy balances for the system. It is con- Using these expressions, the component material balance
venient to write these for each tray in the column and for the (la) can be rewritten as
overhead condenser and for the reboiler. A typical tray in
the system can be represented as shown in Figure 2.
The trays are numbered from the top of the column. The
following assumptions are made: the liquid phase is per- (7)
fectly mixed; there is no mixing of the vapour as it\ passes
through the liquid on a tray, but upon leaving, it is com- or in simpler terms
pletely mixed; the liquid and vapour phases are in thermal
equilibrium at the temperature of the liquid on the tray; the 1, = - A v ~ J ~ l . J +A,v,
l + ...................... (8)
system is at constant pressure; the vapour phase follows where
ideal gas laws; and heat losses from the system are negli-
gible. Using these assumptions, one can derive the fol-
lowing steady state equations. The component material bal-
ance for the jth tray is given by
1ij = Ljxij ; vij = Vj~ij

VJ+ I + I
YIJ + Lj - 1x1~ ~ JLjxij + Fx/ - W'W
- I - V J ~- = 0 Similar expressions for all the trays are combined and solved
........................................ (la) using the 8 method of convergence as described by Holland
The total material balance for the jth tray is given by (198 I). The constants used for calculating the physical prop-
erties, i.e. vapour pressure, liquid enthalpy, and vapour
Vltl + LJ-1 - VJ - LJ + F - W = 0 . . . . . . . . . . (Ib) enthalpy are summarized in Table I .
The enthalpy balance for the j t h tray is given by The key parameter in calculating the vapour and liquid
equilibria for non-ideal mixtures is the activity coefficient
+
VJ+lHJ+I+ LJ-lhJ-l- VJHJ- LJhJ Fh/ - Wh, = 0 y I . Data are given in Hala et al. (1968). To calculate these
........................................ (lc) phase equilibria, several methods including the Wilson

782 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 62, DECEMBER 1984
 TABLEI
Physical Property Correlations
Vapour pressures by Antoine Equation: Log ( P ) = A - B/(T + C)
P = mmHg
T = "C

Antoine coefficients
Margule coefficients Liquid
A B C enthalpies
Acetone 7.240 1279.87 237.50 0.0 0.278 0.741 30.6X IO-'T
Methanol 8.072 1574.99 238.86 0.276 0.0 0.390 19.74XlO-'T
Water 7.967 1668.21 222.00 0.632 0.224 0.0 18.016x 10-'T
Heats of mixing: h,a + bx + cx2 + dr' + ex4 *
Acetone/methanol 1.01722x lo-' -0.753418 0.94408 -0.192395 0
Water/methanol 6.05874X lo-' 0.171893 0.646148 -0.813034 0
Water/acetone -6.00506x lo-' 0.368503 -2.1499 5.30655 -3.50732
*Int. Critical Tables, 5, 155-159.

Method, the Van Larr equations, UNIFAC and others may Then, the equilibrium flash calculations for each tray
be used. Jackson (1974) used the Wilson Method. However, beginning at the reboiler were solved using Newton's Meth-
we found through a number of simulations and comparisons od to determine the equilibrium vapour compositions at each
with literature data, that this method did not given accurate tray. The iterative calculations were stopped when the tem-
predictions over the range of the data available. The 3rd perature changes were less than O.ooOo5.
order Margule method was found to give a more accurate
representation of the data. The form of the equations is Dynamic model
Log ( y l ) = x*J(alJ+ 2xl(aJl - The dyanmic model for the column was formulated by
+ x 2 m ( a i m + 2xi(am - attn)) + X / X m ( a / , + a r m - am]) adding the dynamic terms for the tray holdups and the liquid
holdups in the reboiler and the overhead condenser. Addi-
+ 2X,"(U, - aJm)- c(l - 24)) . . . . . . . . . . . . . (9) tional assumptions were that there was constant volumetric
where the i , j , and rn indicate the components. The sequence liquid holdup and that the fluid dynamic response time was
of calculation is i = 1, j = 2, rn = 3 then i = 2, j = 3, rn negligible compared with that for mass transfer. Based on
= Iand,finally,i=3,j= l,m=2,andc=0.113. these assumptions, the transient equations can be written as
Constants for Equation (9) are given in Table 1. follows:
To solve the steady state equations the sequence of calcu-
lations was as follows. First a composition profile, x,, was
obtained. From this, tray to tray flash calculations were used
to obtain y,, and the temperature profile. Then Equations (2a) + Fx,- WX, . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10)
and (2b) were used to compute liquid and vapour enthalpies.
Finally a new flow profile was calculated from Equation
(Ic). Calculations were repeated until the column profiles
converged.
The steady state equations were solved by fixing the feed After some simple manipulation, the following equations
compositions and flow rates to the column. A linear tem- can be obtained.
perature profile was assumed initially. For the first pass the
reflux rate and the reflux plus solvent, plus feed rates, re-
spectively, were used for the column flows. The distillate
rate was assumed to be the vapour flow rate. The set of
equations resulting from Equation (8) were solved using
Thomas Gedes' Algorithm to get net component liquid flow
rates. '0' is then calculated using the Newton Raphson The Murphree tray efficiencies, Ell calculated from the
Method from which one obtained the corrected flow rate to steady state simulation runs were used in the dynamic sim-
compare with the distillate specification. New liquid com- ulation to calculate the true compositions of the vapour
positions were then calculated using the following relation: leaving each tray. The relationship for the j t h tray was

pi = ( d l ) ~ o / ( d l ) c u y v = YIJ+I + E,(y,* - Y I J + l ). . . . . . . . . . . . . . . . ( 13a)

Since no vapour enters the reboiler, the Murphree effi-
Pm = (dm)co/(dm)cu
ciency was defined differently using the composition of the
N liquid product in place of the vapour in Equation (13a)
xq = (lij)cup,/ C
k=l
(lmj)cuPm
y1.14= xl.14 + E , . I ~ ( Y , . -
I ~X*l , 1 4 ) . . . . . . . . . . . . . . (13b)
ca = calculated The differential equations were solved at each tray for the
liquid and vapour flow rates as a function of time. The
co = corrected using 0. See Holland (1981) for details. enthalpy balance at each tray was defined as
THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 62, DECEMBER 1984 783
 TABLE2
Steady State Operating Conditions (Experimental)
Run number 1 2 3 4 5
Feed
Flow ml/min I09 111 108 I30 108.6
Temp. "C 40.5 40.0 40.7 40.0 40.0
Comp. Acetone 0.637 0.5578 0.637 0.5578 0.629
(mole frac.) Methanol 0.363 0.4422 0.363 0.4422 0.371
Solvent
Flow mt/min 53.5 53.5 53.5 53.5 53.0
Temp. "C 60.4 64.0 60.4 64.0 60.0
Reflux
Flow ml/min 142 I42 I68 146 217.9
Overhead product
Flow ml/min 88.6 81.5 83.3 97. I 80.7
Comp. Acetone 0.81882 0.8002 0.87 I37 0.77295 0.9053
(mole frac.) Methanol 0.15201 0.1712 0.09961 0.18360 0.0572
Water 0.029 1 6 0.02873 0.02902 0.04370 0.0375
Bottom product
Flow ml/min 73.0 81.0 75.6 95 .O 82.2
Comp. Acetone 0.00123 0.00327 0.00191 0.00580 0.00352
(mole frac.) Methanol 0.12550 0.16533 0. I290 0.18571 0.1665
Water 0.87329 0.83233 0.8693 0.80949 0.8299

Experimental and simulation results

The open loop dynamics were investigated by applying

- Wh,, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (14) step changes in feed rate, reflux rate, feed composition, and
reboiler duty, but with the liquid levels in the reboiler and
The change in enthalpy at each time step was assumed to
the overhead condenser under proportional-integral control
vapourize more or less liquid on the tray thus causing a
during these tests. Details of the response of these control
change in the vapour flow rates. The new vapour flow rates loops may be found in Langille (1983).
were calculated from To investigate the steady state behaviour of the column,
a number of experiments were carried out at different oper-
ating conditions. The initial transients from a cold start on
the process required 4-5 hours to settle out although this
The vapour flow rates were adjusted by beginning the calcu- could be reduced by running under total reflux until the trays
lation at the reboiler. The revised liquid flow rates were were fully loaded. The overhead and bottoms compositions
adjusted from the top of the column. for various runs are given in Table 2 with the simulation
This set of dynamic equations was solved using the results in Table 3. The Murphree efficiencies for each tray
DYNSYS Simulation Package with the stiff option in Gear's were adjusted to give the best fit of the data. The water
Method for the integrations. This option was required be- efficiency was set to 1.0 because its K-value did not vary
cause the difference between the reboiler dynamics and the greatly by contrast with that for acetone. The Murphree
tray dynamics is very large. To reduce the number of equa- efficiencies which were finally used in the steady state and
tions in the simulation, the integrations were done on the dynamic simulations are given in Table 4.The reflux ratios
acetone and methanol compositions with the water being from the experimental runs were used directly in the
obtained by difference. At each time step, the vapour com- simulations.
positions for the entire column are recalculated using the Large control actions were required by the reboiler level
bubble point method to give the new equilibrium tem- controller. This resulted in large variations in the bottoms
perature profile. flow rate compared with those for the top product. As a
The stiff option in Gear's method requires that the result, the overhead product rate was used with the mass
Jacobian Matrix for the system be calculated. This was done balance equations to calculate the bottoms product rate. The
by taking partial derivatives of Equation (12) with respect to results showed good agreement with the experimental
x i j - l , xu and other terms as required. observations.
Some problems were encountered with the automatic step An examination of the data in Table 3 shows that the
size adjustment in Gear's Method as the system approached largest errors in the simulation were for the acetone com-
steady state. This resulted in step sizes which were too large positions in the bottoms product stream. This was due in
to allow critical dynamics in the non-linear system to be part to the difficulty in analyzing the bottom composition.
calculated. The automatic step size adjustment was re- The primary reason for this was the resolution of the chro-
stricted by specifying a short time interval for each solution matograph column for the methanol/water system which
and using the results of the previous solution as the initial comprised two polar solvents. The large water peak from the
condition for the next solution. chromatograph dominated the methanol peak and both were

784 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 62, DECEMBER 1984
 TABLE3
Simulated Steady State Operating Conditions and Comparison to Experimental Data
Overhead Bottom Bottom
comp. comp. ACI A Cz flow rate A C3 Run
Acetone 0.81883 0.013703 O.oooO1 -0.012473 73.3 ml/m -0.3 ml/m
Methanol 0.15649 0.123954 +0.00423 0.001546 Run 1
Water 0.02467 0.862341 -0.00449 0.01I929
Acetone 0.8048 0.01834 -0.0046 -0.01507 81.25 ml/m -0.25 ml/m
Methanol 0.1736 0.1667 -0.0024 0.00137 Run 2
Water 0.02 I5 0.81487 +0.0072 0.0I746
Acetone 0.86749 0.0 I927 0.00388 -0.01736 75.6 ml/m 0.0 ml/m
Methanol 0.10929 0.13969 -0.009 -0.0 1069 Run 3
Water 0.02321 0.84 I03 0.005 0.02827
Acetone 0.77807 0.0230 -0.00512 -0.0145 84.8 ml/m +0.2 ml/m
Methanol 0.198479 0.1801 -0.0148 0.00561 Run 4
Water 0.02344 0.7995 0.0202 0.009
Acetone 0.8909 0.0288 0.0144 -0.0193 80.3 ml/m +1.9 ml/m
Methanol 0.0891 0.1564 -0.0319 0.0101 Run 5
Water 0.0199 0.8207 0.0176 0.0092
AC = (experimental-simulated)

TABLE4 TABLE6
Murphree Efficiencies for the Column Relative Volatilities

Tray Acetone Methanol Water Methanol Acetone

1 I .o 1 .o 1.o 1.0505 1.0212

2 I .o 1 .o I .o 1.0304 I .0502
3 1 .o 1.o 1 .o 1 .oo 1.1061
4 I .o I .o 1 .o 0.9680 I .2532
5 1 .o 1 .o 1 .o I .0876 2.0219
6 0.85 1 .o I .o 2.385 5.8783
7 0.85 0.9 I .o 2.7703 5.8783
8 0.85 0.9 I .o 1.9294 4.6713
9 0.85 0.9 I .o 1.7482 4. I793
10 0.85 0.9 I .o 1.4973 3.4767
11 0.95 1 .o I .o I . 1756 2.5290
12 0.95 1 .o 1 .o 1.3174 3.0752
13 0.95 I .o 1.o 1.7492 4.4144
Reboiler 1 .o 1 .o 1 .o 3.4668 8.8966

TABLE5
very large relative to the acetone peak. Small errors in the
Column Temperatures two large peaks had a very large effect on the calculated
acetone composition. With a better method of analysis, this
Tray Liquid Vapour Simulated problem would have been less severe.
The vapour and liquid temperature data corresponding to
1 54.89 - 55.90 Run 2 are shown in Table 5 along with the simulation results
2 55.67 55.38 56.35
56.85 56.50 57.04
for the same case. One can see that the assumption of equi-
3
4 56.967 57.18 58.38 librium conditions is reasonably satisfied. The temeprature
5 63.94 63.57 62.02 data have been plotted in Figure 3 to show the actual profile
6 7 1.03 70.43 72.74 in the column. One can observe the effect of the high tem-
7 69.24 68.89 70.79 perature solvent stream entering the column.
8 68.08 67.59 69.98 Simulated composition profiles matching the temperature
9 67.51 67.02 68.84 profile are also plotted in Figure 3. The methanol com-
10 66.89 66.33 67.08 position in the liquid passes through a maximum corre-
11 63.59 64.113 64.34 sponding to the minimum temperature just below the feed
12 67.155 67.416 66.78 tray. The acetone composition in the vapour phase increases
13 70.69 71.147 71.59
14 Reboiler 81.30 82.08
rapidly in this region and also in the region around the
solvent feed tray. The methanol composition increases
Operating conditions corresponding to Run 2. above this tray both because there is very little water above

THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 62; DECEMBER 1984 785
 85.8,

TEMP PROFILE P ACETONE I N TOP PRODUCT

75 0 ~

n
0
0
78 0 -
W
3 860-
- SIMULATION RESULTS
a 60B-
S 0 EXPERIMENTAL DATA
I-
W 550- 0

--FITTED TRANSFER FUNCTION

8 - 0 IS
ACETONE I N VAPOUR
8 8 8 I1 2
H
u
4 4
MEOH I N LIO LL

C C t L .
./ .- 8
2 w
0
MEOH I N BOTTOM STREAM 0 8 7 LL

005
_I

8
0 3 5 48 4 5 5 8 5 5 1 8 1 5
TIME CHR)

Figure 3 - Column composition and temperature profiles. Figure 5 - Reboiler heat duty step responses.

8 85

883 - + - .a\ ACETONE I N TOP PRODUCT 8'84t

n. 81
.
ACETONE I N TOP PRODUCT

6- 0-
-SIMULATION RESULTS
L
o .EXPERIMENTAL DATA LL 8.76 - SIMULATION RESULTS
W 0 EXPERIMENTAL DATA
--FITTED TRANSFER FUNCTION
B 76 -a 20 8 I1

- 8 18 8 I4

......
.-. a -L -- -
8.l1 E

LL
8 I2

B 18
0
z
I-
u

6- -A MEOH I N EOTTOM STREAM 8.12 'I! MEOH I N BOTTOM PRODUCT 8 88

I?
eJ j I 18 d= W
-1

3 7 44 5 1 5 8 1 5 7 2
800 P
8,E 7.8 8.8 0.8 18.8 II 8
TIMECHR) TIMECHR)

Figure 4 - Feed composition step responses. Figure 6 - Reflux flowrate step responses.

this tray and because the equilibrium coefficient for meth- responses are the acetone in the overhead and methanol in
anol is relatively high as compared to acetone. Due to the the bottoms product compositions.
relatively small increase in relative volatility in the middle Figure 4 shows the responses to a step change in feed
section of the column, the acetone composition shows little composition from 0.637 mole percent acetone to 0.5578
increase here. The presence of water on the trays also pre- mole percent. The acetone composition is well modelled by
vents the increase in vapour composition below the solvent both the empirical transfer function and the simulation.
feed tray. The relative volatility data for the temperature There is a significant transport lag in this response. The
profile shown in Figure 3 are summarized in Table 6. These methanol composition in the bottom follows the direction of
data reflect the temperature effect through the partial the experimental response but it is shifted in time. This is
pressures and the composition effect through the activity attributed to an error in the initial conditions. It is empha-
coefficients. sized here that no parameters have been adjusted in the
dynamic simulation except for the tray holdups which were
Dynamic response of the column estimated to be 400 ml and the reboiler which was 6 litres.
These were fixed for the simulations.
In order to estimate the process time constants and to The responses to a step increase in reboiler heat duty
show the dynamic behaviour of the column, a number of (steam pressure) are shown in Figure 5. The bottoms com-
step tests were performed. Step changes in feed com- position responds very quickly as expected whereas the
position, feed flow rates, reboiler duty, and reflux rates overhead acetone composition shows the rather long lags
were implemented both on the column and in the simulation. due to mass transport in the column. The general shape of
The initial conditions, i.e. column profiles, for the dynamic the response from the simulation is different for the over-
simulation were provided by the steady state program. For head composition although the time to reach equilibrium
each case, the experimental data, the simulation results and agrees well with the data.
a fitted set of transfer function models are shown. The key Figure 6 shows the response of the column to a step

786 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 62, DECEMBER 1984
 TABLE7
Transfer Functions
Manipulated Top product Bottom product
variable (acetone) (methanol)
I . Feed composition 0.53 exp (-7.0s) -0.366 exp (-8.0s)
0.637+ 0.5578acetone 30.3s + 1 45.0s + I
2. Reboiler duty -0.0296 exp (-5.0s) -0.01 exp (-2.0s)
(steam pressure) 400.0~~+ 26.88s + 1 37.0s + 1
22 + 25 psi

3. Reflux flow rate 0.00132exp (-6.0s) 0.00065exp (-17.0s)

149 + 180 ml/rnin 883s2+ 25.43s + 1 52.8s t 1
4. Feed flow rate 0.OOO I 5089 0.00177
109 + 142.8 ml/min 900s2 + 48s + 1 90.0s + I

8 8.8,L ACETONE I N TOP PRODUCT I 8,831 ACETONE I N TOP PRODUCT

LL

-I 8.79
W -SIMULATION RESULT
0

o &EXPERIMENTAL DATA
8.77

-- FITTED
- SIMULATION RESULTS
TRANSFER FUNCTION
18.25 s 8.75
b 4 EXPERIMENTAL DATA

HEOH I N BOTTOM PRODUCT 8 16 2LL

I -- FITTED TRANSFER FUNCTION

MEOH I N BOTTOM STREAM

I
ee 1 8 2 8 3 8 4 8 5.8 0.0 0.5 1.0 1.5 2.0 2.5 9.0 3.6
TIME CHR) TIMECHR)

.Figure 7 - Feed flowrate step responses. Figure 8 - Troubleshooting feed flowrate step responses.

change in reflux flow rate from 149 ml/min to 180 ml/min. itself. The final results was that the reflux temperature con-
The bottom composition shows non-minimum phase behav- troller was malfunctioning with the result that the reflux
iour. The gains for the reflux step are small. The reasons are stream was entering the column at a higher temperature than
that a high reflux rate tends to reduce the solvent concen- that on the top tray. This is similar in effect to increasing the
tration by increasing the non-solvent component in the reboiler heat load and when this was implemented in the
liquid phase. Therefore, the advantage of the higher reflux simulation, the results matched the experimental responses
ratio is reduced due to a concurrent decrease in relative shown in Figure 8. The interesting observation one can
volatilities. The non-minimum phase behaviour is shown make here is that a simulation can be very useful for process
both by the experimental response and by the simulations. troubleshooting. Had we only used fitted transfer functions,
The differences between the two are caused by poor resolu- the problem may not have been diagnosed. It is noted here
tion of our process chromatograph. also that for these multicomponent systems it is much more
The response to a step change in feed flow rate from difficult to predict the effects of changes in the manipulated
109.9 to 143.2 moles/hr is shown in Figure 7. A slight or load variables than for the equivalent binary systems. The
non-minimum phase response is shown for the overhead difficulties arise, for example from system nonlinearities,
composition by both simulation and experiment. The ace- from poor or inadequate measurements (compositions are no
tone Composition rises due to the lower column temperatures longer measureable by tray temperatures), and by operating
(for the same reboiler duty). The simulation does not fit ranges where desired outputs are uncontrollable by the col-
the bottoms composition well. The column responds less umn manipulated variables.
quickly than predicted by the model. Table 7 shows the fitted transfer function models for all
Finally, Figure 8 shows the same column response to a of the step response tests. The models were obtained by
step change in feed flow rate. This run was nominally the non-linear least squares estimates for models of either first
same as for Figure 7 but the response of the process was or second order dead time form. The criterion for which
puzzling since it went in the reverse direction to that of the model to accept was based upon minimization of the sum of
previous results. This was not initially predicted by the squares of residuals. Except for the step response of the
simulation and prompted an investigation of the process overhead product to feed composition which had a large

THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 62, DECEMBER 1984 787
 TABLE8
Steady State Gain (ABS) 10% Disturbance
0 8 - Gain (top) Gain (bottom)
Feed composition
1st Condition
8 6 -
(0.5578, 0.4422) 0.0259 -0.0186
2nd Condition
z
H
+ 0 4 - 0 -
FOR RUN # 2
= FOR RUN # 4
(0.65. 0.35)
Feed flow rate
0.03 I -0.0204
U
lL
A

0 - FOR RUN # 5 1st Condition

W 109.87 moles/hr 0.013597 0.018287
8 2 - 2nd Condition
100 moles/hr 0.02655 0.023754
Refluflow rate
8 8 2 8 4 8 8 8 8.8 180 128 148 I st Condition
PLATE NUMBER 142 ml/min at (inlet feed
of 0.5578, 0.4422) 0.04138 0.012446
Figure 9 - Solvent liquid composition profiles. 2nd Condition
142 ml/min at (inlet feed
of 0.673, 0.363) 0.041492 0.014585
Reboiler duty
1st Condition
1628.I17 kcal/hr 0.0937 -0.048495
ACETOkt IN TOP PRODUCT 2nd Condition
(EXPERIMENTAL DATA) 1709 kcal/hr 0.0846 -0.05
Gain (top) = acetone composition.
Gain (bottom) = methanol composition.

feed points should be as flat as posisble. The profile will

depend upon the solvent flowrate, the feed flowrate, the
feed composition, and the reflux flowrate. More work is
underway to quantify the effects of these variables.
Figure 10 shows the effect of the solvent flowrate. A
severe negative stepchange from 176.0 moles/hr to 24.5
8 8 1 8 2 8 5 8 4 8 5 8 6 8 moles/hr was introduced. From Figure 10 it is clear that the
TIME (HR)
top product composition has gone below 80% clearly indi-
Figure 10 - Solvent flowrate step responses. cating the importance of water in the separation. Also com-
positions lower than 80% indicates that the feed composition
(0.60,0.40)and the operating conditions of the tower do not
allow greater separations.
dead time, all the overhead composition responses are well To illustrate the non-linear behaviour of the column,
represented by second order transfer function models. The steady state gains for 10%disturbances in feed composition,
bottom compositions tend to be first-order dead-time models feed flowrate, reflux flow rate and reboiler duty were calcu-
with rather larger time constants than the above. This lated for two different operating conditions for the column.
probably reflects the large holdup in the reboiler relative to These are summarized in Table 8. The non-linear effects are
the tray holdups. particularly significant for the feed flowrate changes and for
The effect of the solvent to feed ratio can be seen from the feed composition changes. This is an important result
Run 2 and Run 4. In Run 4 the feed ratio is lower giving rise and illustrates the importance of the complete simulation.
to a lower relative change in the volatility of acetone to Fitted transfer functions could only be used to derive control
methanol. The result is that the separation is poor compared schemes in a region for which the implied linearization was
with that in Run 2. This indicates that the solvent flowrate valid. By contrast, the detailed nonlinear models show valid
to feed flowrate ratio has a critical role in controlling the top responses throughout the operating range of this column.
product compositions. To illustrate this point, the solvent Transfer functions derived from these models could be used
liquid composition profile for the column for Runs 2 , 4 , and for control system design purposes.
5 are shown in Figure 9. From these profiles one can see that
the nature of the profile is the same for all cases in that there Conclusions
is a sudden increase in the solvent composition at the solvent
and feed trays, as expected. However, the factor that Steady state and dynamic mass and energy balance
governs the separation is the region between the two feed models have been developed for an extractive distillation
points. The profile steepness declines from Run 4 to Run 2 system separating acetone and methanol using a water sol-
to Run 5 in the reverse order of the acetone separation. From vent. The importance of good simulations for nonlinear
this we conclude that for better separation in extractive multicomponent systems is highlighted. The simulations
distillation the solvent concentration profile between the have been fitted to process operating data by adjusting

788 THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING. VOLUME 62, DECEMBER 1984
 Murphree efficiencies for the trays. Literature data were tillation Column”, Ind. Eng. Chem. Process Des. Dev. 8,
used for the equilibrium calculations for the three- 433-438 (1969).
component mixture. The dynamic models used the results Cho, Y. S . and B. Joseph, “The Development of Simple Dynamic
from the steady state simulations and adjusted only the tray Models. for Separation Processes”, AlChE Meeting, Chicago,
holdups. Finally, transfer function models were fitted to the Illinois, November (1980).
step response tests to allow for simple controller testing. Christiansen, L. J., M. L. Michelsen and A. Fredenslund, 12th
Symp. on Comp. Application in Chemical Eng., Montreaux.
Generally, the simulations gave good agreement with ex- Switzerland, April 8- I I , 1979.
perimental results over a broad range of nonlinear operating Davison, E. J . , “The Interaction of Control Systems in a Binary
conditions. The non-linear behaviour of the column was Distillation Column”, Automatica, 6, 441 -461 (1970).
adequately simulated by the models. These full dynamic Edgar, T. F. and C. 0. Schwanke, “A Review of the Application
models are necessary when reduced order models such as of Modern Control Theory to Distillation”, Proc. Joint Auto-
the bilinear models used by Benallow et a]. (1982) are matic Control Conference, 1370- 1376, San Francisco (1977).
developed. Future work will be focussed on the closed loop Gerester, J. A., Chem. Eng. Prog. 65, 43 ( 1969).
behaviour of the system because the highly nonlinear pro- Hala, E., I. Wichterle, J. Polak and T. Boublik, “Vapour-Liquid
files in the column are difficult to section otherwise. Equilibrium Data at Normal Pressures”, Pergamon Press,
49 1 -494 ( 1968).
Hirose, Y., Y. Kawase and 1. Funda, Ind. Eng. Chem. Process
Acknowledgements Des. Dev. 19, 505-508 (1980).
Holland, C. D., “Fundamentals of Multicomponent Distillation”,
The research reported in this paper was supported by McGraw-Hill, New York (1981).
Operating Grants A6075 and A7017 from the Natural Sci- Jackson, W. D., M. Eng., Chemical Eng., McMaster University,
ences and Engineering Research Council of Canada, by a “Preliminary Design and DDC of an Extractive Distillation
grant from Imperial Oil Limited, and by an equipment grant Column”, July (1974).
from McMaster University. This research support is greatly Langille, K. B., M. Eng. Chemical Eng., McMaster University,
appreciated. “Design, Instrumentation and Commissioning of an Extractive
Distillation Column for Computer Control”, February ( 1983).
Luyben, W. L., Ind. Eng. Chem. Fundam. 14, 321 -325 (1975).
Nomenclature Marr, G. R., “Distillation Column Dynamics: A Suggested Math-
ematical Model”, AlChE Meeting, Baltimore, May 1962.
a,,,, = coefficients for calculating 3rd Order Margules Equation McAvoy, T. and A. Tsogas, “Relative Gain Expression for Dual
= Murphree plate efficiency for plates j and component i
Composition Control”, Proc. Joint Automatic Control Confer-
F = feed flow rate (moles/hr)
ence, Charlottesville, Virginia, WP-7B (198 I ).
HI = vapour enthalpy of the stream leaving plate j (kcal/hr) Osborne, W. G., Jr., D. Reynolds, J . B. West and R. N. Maddox,
hI = liquid enthalpy of the stream leaving plate j (kcal/hr) AIChE, Inst. Chem. Eng. Symp. Ser. 1, “Advances in Sepa-
h*,, = liquid enthalpy of component at plate j
ration Techniques”, 84-89, London (1965).
H*,I = vapour enthalpy of component i at plate j
Patke, N. G. and P. D. Deshpande, Ind. Eng. Chem. Process Des.
k,I = k-value of component i at plate j vapour pressure/total Dev. 21, 266-267 (1982).
pressure
Rosenbrock, H. H., “Distinctive Problem of Process Control”,
Ktj =y#jktj Chem. Eng. Prog. 58, 43-50 (1962).
LI = liquid flow rate leaving plate j Stewart, W. E., R. L. Levien and M. Morari, “A New Model
= component liquid flow rate at plate j for i component
Reduction Technique for Staged Separation Operation”, paper
MI = liquid holdup at plate j presented at the 74th Annual AIChE Meeting, New Orleans, LA
N = number of components in the system ( I98 I).
Q = enthalpy at reboiler ( Q R ) or condenser ( Q c ) Tetlow, N. J., D. M. Groves and C. D. Holland, “Analysis and
V, = vapour flow rate leaving plate j (moles/hr) Control of a Generalized Model of a Distillation Column”, 58th
v , ~ = component vapour flow rate National Meeting, AlChE. Feb. 7 (1966).
W = side stream drawoff rate (moles/hr) Tung, L. S. and T. F. Edgar, “Development and Reduction of a
xY = liquid mole fraction of component i at plate j Multivariable Binary Distillation Column with Tray
yU = vapour mole fraction of component i at plate j Hydraulics”, AlChE Meeting, Houston, Texas, April (1979).
Y * , ~ = vapour in equilibrium with liquid Waller, K. V., “University Research on Dual Composition Control
of Distillation - A Review”, Chemical Process 11, Proceedings
Greek letters of the I98 1 Engineering Foundation Conference, The Cloister,
Sea Island, Georgia, January 19-23, 1981. Edited by Thomas
a = relative volatility
F. Edgar and Dale E. Seborg. Sponsored by AlChE (1982).
yii = activity coefficient of component i at plate j
Waller, K. and S . Gustafsson, “On the General Characteristicsof
Multicomponent Distillation Dynamics”, Kern. Teollisuus,
References 29(3), 173- 184 (1972).
Wong, K. T. and R. Luus, Can. J . Chem. Eng. 58, 382-388
Ballard, D., C. Brosilow and D. Kahn. “Dynamic Simulation of (1980).
Multicomponent Distillation Column”, AlChE Meeting, Miami, Wood, R. K. and M. W. Berry, “Terminal Composition Control
Florida, November ( 1978). of a Binary Distillation Column”, Chem. Eng. Sci. 28,
Benallow, L., D. E. Seborg and D. A. Mellichamp, “Low Order, 1707-1717 (1973).
Physically Lumped Dynamic Models for Distillation Column”,
AlChE Meeting, Los Angeles, California, November (1982).
Berg, L., Chem. Eng. Prog. 65(9), 53 (1969). Manuscript received July 3, 1984; revised manuscript received
Buck, E. and R. R. Bannister, Chem. Eng. Prog. 65(9), 64 (1969). September 27, 1984; accepted for publication September 28. 1984.
Burman, L. K. and R. N. Maddox, “Dynamic Control of Dis-

THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING, VOLUME 62, DECEMBER 1984 789