Pergamon Chemical Engineering Science, Vol. 50, No. 17. pp. 2691 2703.

1995
Copyright © 1995 Elsevier Science Lid
Printed in Great Britain. All rights reserved
0009 2509/95 $9.50 + 0.00
0009-2509(95)00113-1

M U L T I P L E STEADY STATES IN BINARY

DISTILLATION--THEORETICAL AND EXPERIMENTAL
RESULTS

A. KIENLE,* M. GROEBEL and E. D. GILLES

Institut ffir Systemdynamik und Regelungstechnik, Universitfit Stuttgart, Pfaffenwaldring 9,
70550 Stuttgart, Germany

(Received 29 August 1994)

Abstract--Steady-statemultiplicity and instability in distillation is studied theoretically and experimentally

for columns with volumetric reflux. Using nonlinear wave propagation theory for distillation processes, we
derive simple expressions for the prediction of multiplicity regions in parameter space with paper and
pencil. This method also gives further insight into qualitative column dynamics in the region of multiple
steady states. The results are illustrated for ideal binary mixtures and further conclusions are drawn for
nonideal binary mixtures and ideal or nonideal multicomponent mixtures. With regard to experimental
work, a rigorous, numerical bifurcation analysis of more detailed models is presented for the separation of
methanol-l-propanol. The interaction between the boundary set of feasible operation and the bifurcation
set is shown to be of great importance for the understanding of multiplicity and stability features for this
type of column configuration. Finally, these multiple steady states are confirmed experimentally.

1. I N T R O D U C T I O N librium of heterogeneous mixtures found by

The first simulation of multiple steady states in distil-
lation has been published by Magnussen et al. (1979)
for separations involving potentially heterogeneous,
azeotropic mixtures. Subsequent theoretical analysis
seemed to provide some evidence that multiple steady
states in distillation are exclusively confined to separ-
ations involving mixtures with a potential phase split
of the liquid phase. In particular, Doherty and
Perkins (1982) proved uniqueness and stability for
distillation involving binary, homogeneous mixtures.
Their analysis has been built on a constant molar
overflow model, which neglects the influence of the
energy balance. Further, Sridhar and Lucia (1989)
proved that this result is not affected by the energy
balance for two different specifications of the input
variables. The generalization of these results to multi-
component mixtures, however, has been limited to the
theoretical case of fixed temperature and pressure
profiles (Sridhar and Lucia, 1990). The models used in
Doherty and Perkins (1982) as well as in Sridhar and
Lucia (1990) were on a pure molar basis.
Nowadays, it is a widely accepted fact that multiple
steady states in distillation can have different sources
and that multiplicity is also possible in homogeneous
and even in binary distillation. So far, the following
sources for multiple steady states in distillation col-
umns have been found:
(i) Cairns and Furzer (1990) showed that multiple,
metastable solutions of the vapour liquid equi-
Van Dongen et al. (1983) are a potential source
for spurious multipli-cities in the simulation of
heterogeneous azeotropic distillation columns.
(ii) Instead, Jacobsen and Skogestad (1991b)
showed that physical reasonable multiplicities
can even occur in ideal two-product distilla-
tion. They found two different, potential
sources for this type of behaviour.
(iil) The first is a nonlinear input transforma-
tion for columns with mass or volume
inputs (e.g. volumetric reflux and molar
boilup).
(ii2) The second is the presence of the energy
balance for different sets of specifications
than considered by Sridhar and Lucia
(1990).
(iii) More recently, Morari and coworkers
(Laroche et al., 1992; Bekiaris et aL, 1993b)
and independently Kienle et al. (1992, 1993)
found also multiple steady states in homo-
geneous azeotropic distillation with constant
molar overflows. The models were again on
a pure molar basis. The sources for this type of
multiplicity is clearly seen to lie within the
properties of the vapour liquid equilibrium of
homogeneous azeotropic multicomponent
mixtures. Further, it can be shown that the
same type of mechanism also occurs in hetero-
geneous azeotropic distillation (Bekiaris et al.,
1993a).

*Corresponding author. E-mail: KIENLE@ISR.UNI- Note, that the term "multiple steady states" refers
STUTTGARTD400DE. to output multiplicities, i.e. the concentration and
2691
2692 A. K1ENLE et al.
temperature profiles and hence also the product com-
positions are ambiguous for fixed design parameters
and fixed input variables. This should not be mixed
with the so-called input multiplicities, where the same
product compositions can be achieved with different
sets of input variables and/or design parameters.
So far, all investigations Cited above are of purely
theoretical nature. Nevertheless, the results cited in
item (ii) and (iii) are motivated by physical consider-
ations, which clearly show that these multiplicities
should also occur in real distillation plants. But an
experimental validation of multiple steady states in
distillation is still missing. This gap will be closed in
the present paper. As a first step towards the exPeri-
mental verification of multiplicity phenomena in dis-
tillation we focus on multiplicities of type (iil).
Jacobsen a n d Skogestad (1991b) have shown that
uniqueness and stability of steady states in distillation
depends crucially on the choice of the input variables.
As shown in Fig. 1 possible input variables are the
distillate flow D, the bottoms flow B, the reflux L and
the vapour flow V or alternatively the heating rate
Q of the reboiler. The vapour at the top is assumed to
be totally condensed. Since the column has two de-
grees of freedom at steady state, two of the four
possible input variables can be specified independ-
ently. The remaining two follow from the specified
inputs under steady-state conditions.
In theory, L and V are usually specified as molar
flow rates and--as already mentioned--this so-called
L/V-configuration always has a unique and stable
steady state for homogeneous binary distillation with.
constant molar overflows (Doherty and Perkins,
1982). In practice, however, liquid flow rates are usu-
ally fixed on a volume basis by a corresponding valve.
Instead, the vapour flow rate can be adjusted by the
heating rate of the reboiler in good approximation as
a molar flow rate as discussed in Jacobsen and
Skogestad (1991b).
Let Lq for example denote the volumetric reflux.
The molar reflux L follows from a given volumetric
L D
F, ZF
Fig. l. Column configuration for a two-product distillation
column,
reflux Lq in dependence of the molar volume v(xo) of
the distillate according to
L= Lq (1)
v(xo)"
With respect to eq. (1) a simple block diagram for the
Lq/V-configuration can be drawn as shown in Fig. 2.
Note, that each single block of this diagram has
a Unique solution. Multiplicity and instability is there-
fore a matter of the feedback shown in Fig. 2. In the
remainder, we will study this effect in more detail.
The paper is organized in the following way: In the
first part we present theoretical results. Using nonlin-
ear wave propagation theory for distillation pro-
cesses, we derive simple expressions for the prediction
of multiplicity and instability regions. The results are
illustrated for the binary mixture methanol-l-pro-
panol, which behaves thermodynamically nearly
ideal. Further we draw some general conclusions for
nonideal binary mixtures and for ideal as well as
nonideal multicomponent mixtures. Afterwards a
rigorous numerical bifurcation analysis of more de-
tailed models is presented. The bifurcation behaviour
is surprisingly complex due to the presence of bound-
aries in parameter space, which interact with the bi-
furcation set. Although, the results of the simple wave
model are qualitatively in good agreement compared
to the rigorous model, the numerical analysis turns
out to be a necessary prerequisite for efficient experi-
mental verification of steady-state multiplicities due
to quantitative differences. Finally, we present for the
first time experimental results on steady-state multi-
plicity in distillation. These experimental results are
shown again to be in good agreement with the theor-
etical results.
2. THEORETICALRESULTS
The following is based on nonlinear wave propaga-
tion theory for distillation processes (Marquardt,
1988, 1990; Hwang, 1991). According to this theory
the spatial concentration and temperature profiles in
high purity distillation columns can be interpreted as
travelling waves. The profiles show steep fronts cor-
responding to the regions of intensive mass transfer.
After a disturbance these fronts move with approxi-
mately constant velocity and shape.
In the remainder, we will show that nonlinear wave
propagation theory is not only useful for the under-
standing of qualitative column dynamics including
L- ~
- v(zn) Column D
V L/V-configuration ZB
,zF ?_1 I B
Fig. 2. Block diagram representation of the L~/V-configura-
tion.
 Multiple steady states in binary distillation
multiplicities and instabilities but also allows simple
and explicit formulae to be derived for the determina-
tion of the multiplicity regions in parameter space.
Exemplarily, we will treat the Lq/V-configuration.
However, the approach presented here is easily ap-
plied to other configurations with multiple steady
states, which have been discussed by Jacobsen and
Skogestad (1991 b),
Since multiple steady states for the Lq/V-configura-
tion are caused by the feedback at the top of the
column as shown in Fig. 2, we focus in a first step on
a pure rectifying column as shown in Fig. 3.
2.1. Rectifying column
The column profiles in high purity, binary distilla-
tion with constant molar overflows usually consist of
a single constant pattern wave within each column
section (Marquardt, 1988; Hwang, 1991). Let us for
example consider the composition profile of the more
volatile component in a pure rectifying column. It
represents a single constant pattern wave, converging
monotonically to the pure light component at the top
and to the liquid composition x*(yF), which is in
equilibrium with the vapour feed, at the bottom as
illustrated for three different types of profiles in Fig. 3.
Assuming ideally constant shape, we can derive the
propagation velocities for the different types of pro-
files as follows:
(i) First, we consider a wave close to the upper
boundary (type 1 in Fig. 3). For sufficiently
long columns we can neglect the influence of
the lower boundary and consider the wave to
be ideally pinched at the bottom. This implies
that the liquid is assumed to be in equilibrium
with the vapour feed at the bottom. At the top
we consider a total condenser with negligible
holdup. Under these assumptions the propaga-
tion velocity w follows from a global material
D , XD
i
.H
¢-
0 positive in the vicinity of the intersection point, cor-
0 x;(ur)
x [-]
L, xo
Fig. 3. Different types of composition profiles for a rectifying
column.
CES SO-17-B
2693
balance across the wave according to
A(1 + w) Xo-- yr
A 1 -Ae~ xD - x~(yv)' (2)
Here, x, y are the mole fractions of the light
component in the liquid and the vapour phase,
respectively, and A, e are the ratios of the molar
flow rates and holdups in both phases. For the
details we refer to Marquardt (1988). A similar
approach has been presented by Hwang (1991).
(ii) Second, we consider a wave close to the lower
boundary (type 2 in Fig. 3) and neglect the
influence of the upper boundary corresponding
to an ideally pinched wave at the top. For type
2 waves the global material balance reads:
A(1 + w) 1 - y~
A = - - (3)
1 --A~w 1 - Xo"
(iii) Third, we consider a wave in the middle of the
rectifying column and neglect the influence of
both boundaries corresponding to a wave
ideally pinched on both sides. The correspond-
ing velocity
A(I + w) 1 -- Yv
A (4)
1 --Aew 1 - x*(yr)
can be viewed as an eigenvelocity of such an
ideal wave. In particular, a so-called balanced
wave is achieved for vanishing eigenvelocity
according to
1 --Yv
A(w = 0) = A - (5)
1 - x~tyF)
For the L/V-configuration the flow rate ratio A is
directly fixed by the input variables L, V and does
therefore not depend on the position of the wave.
Hence, if we consider steady state as the limit of
a dynamic transient, with a wave standing initially in
the middle of the column, a steady state of type 1 is
achieved for positive eigenvelocity and a steady state
of type 2 for negative eigenvelocity. The balanced
wave according to eq. (5) therefore represents a criti-
cal boundary in the A, YF parameter plane, which
separates steady states of types 1 and 2. This is illus-
trated exemplarily for methanol-l-propanol in Fig. 4.
For the calculation of the vapour-liquid equilibrium
a model with constant relative volatility is used as
given in Table 1.
The wave velocities for types 1 and 2 profiles ac-
cording to eqs (2) and (3) are shown in Fig. 5 as
functions of the unknown boundary compositions
xo, Xo. It is easily shown that these curves intersect
the axis w = 0 precisely once corresponding to
a unique steady state. Further, the slope is always
responding to a stable steady state. This can be ration-
alized in the following way: Since the column profiles
are monotonically increasing for the more volatile
component, the compositions at the lower and the
upper boundary xD and Xo decrease for a positive
2694 A. KIENLEet aL
For the L~/V configuration the flow rate ratio A de-
pends on the distillate composition according to

A= L4 (6)
Vv(x.)
R
and therefore--in particulate for type 1 profiles--
also on the position of the wave front. As a conse-
quence, simultaneous type 1 and type 2 steady-state
profiles are possible for the same set of operating
0
0 1 conditions. This is illustrated in Fig. 7 for a meth-
A=L/V anol-l-propanol rectifying column. The correspond-
Fig. 4. Regions with different types of wave solutions in the ing operating conditions are given in Table 2. De-
A, y~ parameter space. pending on the initial position of the wave the front
moves either towards a steady state at the upper
boundary (type 1) or towards a steady state at the
Table 1. Physical properties of lower boundary (type 2). In between there is a critical
methanol-l-propanol front position, which separates the domains of attrac-
tion of the two stable steady states. We will show, that
3,55
vl 0.041 m3/kmol this separatrix is the intermediate unstable steady-
v2 0.075 m3/kmol state indicated by the dashed line in Fig. 7.
For type 2, profile XDis equal to one and is in good
approximation constant for small displacements of
the wave front. Consequently, the flow rate ratio A ac-
displacement of the corresponding steady-state profile cording to eq. (6) is also constant and follows for the
and increase for a negative displacement as shown in Lq/V-configuration from
Fig. 6. Moreover, the velocities shown in Fig. 5 are
positive for increasing xD and Xo corresponding to A = Lq (7)
Vvl "
a negative displacement of the steady-state profile and
negative for a positive displacement. Hence, the pro- In other words, the feedback shown in Fig. 2 is pas-
files always return to steady-state position and are sive. Hence, steady-state profiles of type 2 are
therefore stable. Note, that the quasistatic stability completely analogous for the L/V- and Lq/V-
consideration employed here is necessary but not suf- configuration. In particular, they occur in both cases
ficient for dynamic stability, since we neglected for operating conditions corresponding to region II in
changes of the profile shape. On the other hand, Fig. 4. Further, the propagation velocity is in both
quasistatic instability as introduced above is sufficient cases as shown in Fig. 5. Therefore, type 2 profiles are
but not necessary for dynamic instability. Moreover, stable for both configurations.
it provides a good description of the mechanism of Instead, for type 1 profiles the distillate composi-
instability we are concerned with here as will be tion and therefore also the flow rate ratio A in eq. (6)
shown in the remainder. A rigorous stability proof for change with the front position. Hence, the feedback
the L/V-configuration with constant molar overflows shown in Fig. 2 is active. The propagation velocity for
has been given by Doherty and Perkins (1982), as type 1 profiles follows for the L~/V-configuration from
already mentioned above. eqs (2) and (6). Under steady-state conditions this

region I: y F = 0.3, A=0.3 region I1: y F = 0.8, A=0.8

2 2
Z

1 1

"T' ,-;.,
0 ~0
-1 . . . . . . . . . . . . . . . . . . . . . . .
-1

-2 -2
0 0.5 0 0.5
x D [-] x_O [-]
Fig. 5. Types 1 and 2 velocities for the different regions in Fig. 4.
 Multiple steady states in binary distillation 2695

XD With respect to this relation eq. (8) represents a quad-

I ratic equation for the unknown steady-state distillate
composition. It is easily shown, that eq. (8) has either
two positive real solutions greater than x~(yD or
none. In particular, the number of real solutions

[
changes from two to zero as the discriminant of eq. (8)
vanishes. The set of all operating conditions where
this happens is the bifurcation set and represents
a curve in the parameter plane. This curve is illus-
trated by the dashed line for methanol-l-propanol in
Fig. 8. For better comparison with Fig. 4 we use the

o!I
normalized flow rate ratio A = Lq/Vvl as the second
parameter in Fig. 8. The molar volumes for meth-
anol-l-propanol are given in Table 1.
Further, the number of physically relevant solu-
0 tions for x o changes by one, if one of the solutions
x [-] z0 I
becomes greater than one. The corresponding bound-
ary is given by
Fig. 6. Displacements of composition profiles of types l
and 2. Lq 1 -- yf
0o)
Vvl 1 - x*(yF)

This equation also represents a curve in the ,4, Yv-

parameter plane, which coincides with the curve of the
balanced wave according to eqs (5) and (7). Together
with the bifurcation set this curve divides the para-
meter plane into four regions as shown in Fig. 8. The
I
corresponding propagation velocities for type 1 pro-
files are shown in Fig. 9. Physically relevant multipli-
._m
cities only occur in region IIa. In summary, we find for
the Lq/V-configuration:

(i) In region I of Fig. 8, we have again a unique,

stable steady state of type 1.
(ii) In region IIa, we have three steady states. One
0 stable steady state of type 2 and two steady
0 states of type 1, one is stable corresponding to
x [-] the positive slope in Fig. 9 and one is unstable
corresponding to the negative slope in Fig. 9.
Fig. 7. Multiple steady-state solutions for the methanol-1- That is, for positive displacements from the
propanol rectifying column. ( - - ) Stable; ( - - - ) un-
stable. unstable steady state the velocity is positive
and the front moves up towards the stable type
1 profile. Instead, for negative displacements
the velocity is also negative and the front
Table 2. Operating conditions for the
methanol-l-propanol rectifying column in moves down towards the stable type 2 profile.
Fig. 7

N 20
Yv 0.3 (saturated vapour)
V 108 mol/h
Lq 3.78 l/h

relation reads

A(w=0)= L~ x o - yv (8)
Vv(xo) x o - x*(yF)"
0
o 1
For an ideal binary mixture the molar volume is given A=Lq/VIv I
by
Fig. 8. Regions with different types of wave solutions in the
v(xo) = v l x ~ + v2(1 - Xn). (9) A, Ye parameter space.
2696 A. KIENLEet al.
region I: y_F =0.3, ~ =0.7 region Ila: y F =0.3, ~ =0.8
0.5 0.5

"-,..
0 0L

S
Z 77

-0.5 -0.5

-1 -1
0 0.5 1.5 2 0 0.5 1.5 2
x_D [-] x_D [-]
region lib: y_F =0.8, ~, =0.47 region IIc: y_F =0.8, ,~, =0.7
0.5 0.5

0 0
77

-0.5 -0,~

-1 -1
0 0:5 1 1.5 2 0 0.5 1.5 2
x_D [-] x_D [-]

Fig. 9. Type 1 velocities for the different regions in Fig. 8.

(iii) And finally, in regions lib, IIc we find a single 0.5

stable steady-state solution of type 2.

F o r the prediction of the multiplicity region we

only had to assume that the column is sufficiently long
corresponding to a pinch at the bottom or the top.
This approximation, however, yields also good results
for comparably short columns as shown in Fig. 10.
Figure 10 gives a comparison between multiplicity
regions predicted by the wave model (N = ~ ) and
a rigorous tray-to-tray equilibrium model with con-
stant molar overflows and different numbers of trays
(N = 5, 10). It is shown that the regions predicted by
the rigorous model converge for increasing N rapidly
to the region predicted by the wave model. In particu-
lar, for N = 10 the multiplicity regions are almost
congruent. But the rate of convergence also depends
on the separation factors and is slower for mixtures
with smaller separation factors and therefore less
steep concentration fronts. Note, that the constant
shape assumption does not affect steady states and
multiplicity regions but only dynamics.
For nonideal binary mixtures the molar volume
follows from
v = 5t (xo, p, T ) x ~ + 52(xa, p, T)(1 -- xD)
where the bar denotes the partial molar volumes,
which are in general also functions of the composi-
tion. Hence, for highly nonideal mixtures eq. (8) is of
higher order and additional steady-state solutions of
type 1 are possible.
N=5
N=10
~lll N=oo
0.5 1.0
,~=Lq/V/v 1
Fig. 10. Comparison of multiplicity regions between the
wave model (N = oo) and a rigorous tray-to-tray model with
(N = 5, 10).
For ideal and nonideal multicomponent mixtures
we have at least Nc - 1 different waves not necessarily
of the constant pattern wave type. Therefore, addi-
tional types of steady states are possible. But for type
1 profiles, where all waves are close to the upper
boundary, we can apply eq. (8) in vector form. Since
this is a system of quadratic or even higher-order
(11)
equations, additional steady-state solutions are also
possible in the multicomponent case.
Finally, the approach presented here is easily ap-
plied to similar configurations with multiple steady
states, which have been discussed in Jacobsen and
Skogestad (1991b).
 Multiple steady states in binary distillation
2.2. Distillation column
We can proceed in a similar way for a distillation
column with rectifying and stripping section. That is
similar, global material balances for the stripping sec-
tion can be derived as done previously for the rectify-
ing section. Afterwards these expressions can be used
to eliminate the unknown composition of the vapour
entering the rectifying section and the composition of
the liquid entering the stripping section. Since the flow
rate ratios in both column sections can be manip-
ulated independently all combinations of types 1 and
2 waves in the rectifying and the stripping section are
possible. This yields four different combinations cor-
responding to four different regions in parameter
space for the L/V-configuration and seven different
regions for the Lq/V-configuration. Because this
brings only little new insight, we will not carry out this
program here in detail. Note, that the qualitative be-
haviour is the same as for the pure rectifying column.
Only size and location of the multiplicity regions
differ due to the interaction between rectifying and
stripping column.
Instead, a numerical bifurcation analysis is used to
study the multiplicity regions for the distillation col-
umn in detail. This turned out to be an important step
towards experimental verification. With regard to ex-
perimental work a distillation column with 40 trays
and feed conditions as given in Table 3 is considered.
The mixture under consideration is methanol-l-pro-
panol. Main points of interest are the limits of feasible
operation not considered explicitly, so far, and the
influence of more detailed models on the multiplicity
regions.
For the experimental investigations we are mainly
interested in the volumetric reflux L~ and the reboiler
duty Q as independent parameters, since these quant-
ities can be manipulated directly at the real plant. It
turns out that the bounds of feasible operation play an
important role in the understanding of the multipli-
city and stability features of this configuration. These
boundaries are represented either by the limit of total
reboil, where the bottom flow rate vanishes, or by the
limit of total reflux, where the distillate flow rate
becomes zero. For simplicity, we will call the set of all
operating conditions where one or both will happen
the boundary set. A theoretical prediction of this
boundary set together with the bifurcation set in the
Lq, Q parameter plane are shown in Fig. 11 for the 40
tray column. Again, the computations shown in
Fig. 11 are based on a tray-to-tray equilibrium model
with constant molar overflows. Typically, the number
of feasible steady state solutions in Fig, 11 changes by
Table 3. Feed conditions for
the methanol-l-propanol
column
F 7 l/h
z~ 0.3
Tr 343 K
2697
6
Unique, stable
steady state
5 -I Uniq
steady state
~ 4 -
~ 3
"~ .~
~ r ~ ! i ':::~: Two steady states
(one stable, one unstable)
2 - ~,...~:~ ::~!~
I Three steady states
, ~!~:Jii::!~.-:~:!ii~' -~~ (two stable, one unstable)
iiiiiiiiiiiiiiiiiij
I .........................
0 I0 20 30
Volumetric reflux Lq (l/h)
Fig. 11. Bifurcation and boundary set for an equilibrium
model with constant molar overflows.
two as the bifurcation set is crossed and by one as the
boundary set is crossed (Balakotaiah and Luss, 1984;
Golubitsky and Schaeffer, 1985). Note, that in the
multiplicity region feasible steady-state solutions are
possible "beyond" the boundaries of total reflux and
reboil as shown in Fig. 11, because the boundaries are
curves running on a folded surface, which represents
the steady-state solution manifold. As the boundary
set is crossed only one solution becomes unfeasible,
but not all. This is illustrated in more detail in Fig. 12
with several cross sections through the solution sur-
face in the direction of Lq.
From Fig. I1 we draw the important conclusion
that the multiplicity region within the cusp grows for
increasing reflux and increasing heating power corres-
ponding to an increasing vapour flow rate. This be-
haviour has also been reported by Jacobsen and
Skogestad (1991b). But an opposing influence is intro-
duced by the boundary set, which captures the region
with three feasible steady states, two of which are
stable and the intermediate is unstable, and cuts off
parts with either two steady states, one stable the
other unstable, or even a single unstable steady state
solution. In the latter region, reasonable column op-
eration is only possible with suitable control. With
respect to experimental investigations, we are parti-
cularly interested in the region with three feasible
coexisting steady states.
The results shown in Figs 11 and 12 have been
obtained with the newly implemented methods for
stability and bifurcation analysis in DIVA (Kienle et
al., 1994), These tools allow a systematic and efficient
investigation of multiplicity and stability features of
distillation columns with different models already
contained in the simulators model library. This turns
out to be indispensable with respect to experimental
investigations.
The corresponding stability diagram for the same
column with the same feed specifications but a more
detailed model is shown in Fig. 13. This model in-
cludes additionally an energy balance, finite mass
2698 A. KIENLEet al.
Q = 2.5kW Q = 3.5kW
1.2

,~.0.8 zo.8)
x 10.6

0.4

0.2
6
Lq [I/h]
Q = 4.5kW
8 10
0.4

0.2
8
I 10
Lq [I/h]
Q = 5.5kW
12 14

1.2 1.2

1 1

~.0.8 .~0.8
C3
×10.6 x 10.6

0.4 0.4

O" 0.2
'"12 14 16 18 16 18 20 22
Lq [I/h] Lq [I/h]

Fig. 12. Bifurcation diagrams for different values of Q corresponding to Fig. 11. ([]) Total reboil; (o) total
reflux.

61 m
verification of multiple steady states in the remainder.
Unique, stable It is worth noting that no additional multiplicities due
steady state
to the energy balance arise for the operating condi-
5 ~ Unique, unstable tions we consider here. For more details on this issue
steady state
we refer to Jacobsen and Skogestad (1991b).

_ ~ ~¶.:"! ¢ ~..~,: f : ~ " 3. EXPERIMENTAL RESULTS

m Two steady states
~iiii~i~i!i~bl(~, one unstable
2 r~..iiiiii~i~, m Threest~iystates
:::::::::::::::::)::::~ ~ ( t : : stable, one unstable
:i))iiiiiiiiiii)iP
~ t I
0 I0 20
Volumetric reflux Lq (I/h)
Fig. 13. Bifurcation and boundary set for the detailed model.
transfer resistance, finite pressure drop, etc. Unknown
model parameters like mass transfer coefficients are
fitted to the real plant. Our experience shows that this
model fits the behaviour of the real column quite well.
Although, the multiplicity regions in Figs 11 and 13
are qualitatively similar, they differ quantitatively in
size. In particular, the region with three steady states
becomes narrower but extends even to higher values
of the heating rate compared to Fig. 11. This is an
important feature we will utilize for experimental
Experimental results have been obtained for the
separation of methanol-l-propanol with a pilot plant
distillation column. The glass column has a diameter
of 100 mm and a height of about 7 m. It is equipped
with 40 bubble cap trays. The column is operated
30 under atmospheric pressure and has a total condenser
at the top. The reboiler is heated electrically. To
ensure constant feed temperature, the feed is pre-
heated before it is fed to the column on tray 21. The
temperature is measured on each tray of the column.
An industrial distributed control system (DCS) PLS
80 from IC Eckhardt is used for the registration of
about 100 analogue and 200 binary signals. About 30
basic control loops for level, flow or heater control are
realized on the PLS 80 system. Automatic column
startup is possible with the dynamic simulator DIVA,
which runs on a computer connected to the DCS.
Moreover, the DCS allows flexible configuration of
the desired column input variables, which is an impor-
tant feature with respect to the experiments discussed
subsequently.
 Multiple steady states in binary distillation
The heating power of the reboiler Q is selected as
the first input variable. From the theoretical point of
view, Q is to be fixed at a value, where the interval
with three coexisting steady states is as large as pos-
sible (see Fig. 13). From the practical point of view,
the heating power is limited with respect to fluid
dynamic stability, which is a necessary prerequisite for
reasonable column operation and has not been taken
into account in the theoretical considerations so far.
Both issues are satisfied for a heating power of
Q = 4 kW in Fig. 13. To reproduce the theoretical
results satisfactory, slightly higher values of the heat-
ing power were required for the real plant compared
to the computations. This is due to heat losses and
other effects like subcooling of the condensate, which
are not taken into account in the models we con-
sidered so far.
In a first step, the heating power is fixed at a value
of 4.5 kW and the external reflux ratio defined by
L for the same value of the input variable L~. Further-
r = --
D more, the intermediate solution branch becomes un-
is taken as the second input variable. The simplified
flowsheet for this configuration is shown in Fig. 14.
Since the distillate flow is considerably smaller then
the reflux flow, it is particularly important for this
configuration to take the reflux flow for level control
of the accumulator at the top of the column and not
2699
the distillate flow. During the startup operation the
product flows are supplied to the waste tank. After the
first operating point has been reached, the product
flows are recycled into the feed tank. The feed condi-
tions for all experiments are given in Table 3.
F o r the configuration in Fig. 14, the reflux ratio r is
fixed at different values and the corresponding vol-
umetric reflux Lq is measured, which is obtained un-
der steady-state conditions after a couple of hours.
The mean measured values are shown in comparison
to the theoretical predictions in Fig. 15. From Fig. 15
we conclude: For any fixed reflux ratio r as input
variable a unique volumetric reflux Lq is obtained as
output variable corresponding to a unique and stable
steady state of the column. But the same value of
Lq can be achieved with three different values of the
reflux ratio r. Therefore, if we switch to the volumetric
reflux Lq as input variable, three different values of the
reflux ratio r corresponding to three different steady
states with three different product compositions exist
(12)
stable.
In a second step, we therefore switched to the
Lq/Q-configuration, to operate the column directly in
the region of multiple steady states. The simplified
flowsheet for this second configuration is shown in
Fig. 16. Note, that the product flows are not recycled
for this type of experiment. The distillate flow is now

r=L/D

40

O-

()

tank2 [~
I

Fig. 14. Simplified flowsheet for the r/Q-configuration.

2700 A. KIENLEet al.
16.5 i - - the tray, where the sensor is located. The lower dia-
- c o m p u t e d values gram in Fig. 17 shows the corresponding measured
o measured values
input variables Lq and Q. Feed conditions are again as
16
given in Table 3.
After column startup not shown in Fig. 17 the up-
x15.5 per stable steady state is reached for a value of the
volumetric reflux of 15 1/h. The corresponding meas-
ured, spatial temperature profile for t = 5000s is
"~ 15
shown in Fig. 18. The dominating front in Fig. 18 is
located at the top of the column. At t = tl in Fig. 17
14.~ the reflux rate is increased in order to push the profile
o towards the second stable steady state, which corres-
ponds to a front location at the bottom of the column.
14
After stepwise reduction of the disturbance Lq is
reflux ratio r = L/D [-]
finally rested to its prior value of 15 l/h at t = t2 in
Fig. 15. Comparison between theoretical and experimental Fig. 17. However, since the system has already passed
results for the r/Q-configuration. the separatrix, which is the intermediate unstable
steady state, it does not return to its prior steady state
but the front keeps on moving towards the second
used for level control of the accumulator at the top, stable steady state, corresponding to a front location
since the volumetric reflux Lq is to be fixed at a pre- at the bottom of the column. The corresponding
scribed value. steady-state temperature profile for t---50,000 s is
The results of this experiment are shown in Fig. 17. shown in Fig. 18 again.
The upper diagram in Fig. 17 shows the time plots of At t = t3 in Fig. 17 the reflux is decreased to push
several representative temperatures Tk, measured at the front towards the upper stable steady state again.
different positions along the column. Index k denotes At t = t4 the reflux Lq is reset to its prior value of

21

40

Fig. 16. Simplified flowsheet for the Lq/Q-configuration.

 Multiple steady states in binary distillation 2701

T, T,o T2o T3o T4o

380 , -- -

S ::
•~ 360 . . . . ' ...........

~ 350 . . . . . . . . . . . .

340- , . . . . . . . . .

330 . . . - . -
0 10000 20000 30000 40000 50000 60000 70000 80000
time [s]

20 :
T
16-:---:-- ......................................... [-
, I[;, , ,v -y=~

"~ 12- -: : ?
0

8- -: I
Q
4- ' I-

0 (

0 20000 30000 40000 70000 80000

time [s]

tl t2 t3 t4 t5

Fig. 17. Measured input and output variables for the L~/Q-configuration.

50000s t = t 5000s
- -,.,.~.~ / 1

5_'~ s -"f--._ - . . 1\
7- 7
9-r 9
11- 11
13- 13

L
15- 15
17- 17
.~ 1 9 - 19 ~%,
J
E 20-
,~ 2 2 - ,- 22
~24- t ~" 24
~ 26
28
30 30
32
34 \ 32
34
'.,,.._ 36
38 ~
40 ~ 1
335 345 355 365 375 335 345 355 365 375
T [K] T [K]

Fig. 18. Measured multiple steady-state temperature pro- Fig. 19. Computed multiple steady-state temperature pro-
files. files. ( - - ) Stable; ( - - - ) unstable.

CES 5 0 - 1 7 - C
2702 A. KIENLEet al.
15 1/h. However, since the front is still in the domain
of attraction of the lower steady state, it returns spon-
taneously and tends towards the lower steady state
again. This return can be seen best for T1 the temper-
ature at the top of the column. Hence, the reflux is
decreased once more again to push the front into the
domain of attraction of the upper steady state. This is
finally accomplished at t = t5 in Fig. 17. After reset-
ting Lq to 15 l/h, the system returns to the first steady
state again. The corresponding temperature profile at
t = 78, 000 s shown in Fig. 18 is nearly identical to the
profile obtained at the beginning. Further, the meas-
ured profiles are in good agreement with the
simulated profiles in Fig. 19. In contrast to Fig. 18 the
dashed profile in Fig. 19 represents the intermediate
unstable steady state, which has not been measured.
The results discussed above have been confirmed in
several experiments,
4. DISCUSSION
For the first time experimental results on steady-
state multiplicity in distillation are presented. A rigor-
ous bifurcation analysis with models of increasing
complexity has been a necessary prerequisite for ex-
perimental verification. Further, it has been shown
that nonlinear wave propagation theory for distilla-
tion process provides a useful and comparably simple
approach to the understanding of these phenomena.
Our experience shows, that the multiplicity regions
for the pilot plant distillation column and the mixture
considered in this paper are relatively small. There-
fore, precise adjustment and control of the input vari-
ables are required for reproducible measurement of
these multiple steady states. In addition, careful col-
umn operation is required for startup operation and
the transient between different, coexisting stable
steady states, since the column is either close to total
reflux or close to total reboil for the operating condi-
tions considered above. Usually, the column con-
sidered here is operated with considerably lower inter-
nal flow rates corresponding to a unique and stable
steady state. But this can be different for other col-
umns and other mixtures.
In practice, the kind of multiplicity and instability
considered here can be avoided either by:
(i) Selecting a different set of input variables, such
as the reflux ratio r instead of the volumetric
reflux Lq as shown in Section 3.
(ii) Or fixing the outputs instead of the inputs by
suitable control of at least one product com-
position (one point control) or a temperature on
a sensible tray, instead. Implications of open
loop instability and multiplicity on closed loop
dynamics have been treated by Jacobsen and
Skogestad (1991a).
Acknowledgement--The financial support of the DFG
(Deutsche Forschungsgemeinschaft) is appreciated.
NOTATION
A molar flow rate ratio ( = L/V), dimension-
less
/T normalized flow rate ratio for the Lq/V-con-
figuration ( = L~/Vvl), dimensionless
B bottom flow rate, kmol/s
D distillate flow rate, kmol/s
F feed rate, kmol/s
L liquid flow rate, kmol/s
Lq volumetric liquid flow rate, ma/s
N number of trays
Nc number of components
Q heating rate
r reflux ratio ( = L/D), dimensionless
t time, s
T temperature, K
v molar volume of the mixture, ma/kmol
v~ molar volume of pure component i,
m3/kmol
partial molar volume of component i,
m3/kmol
V vapour flow rate, kmol/s
w propagation velocity, dimensionless
x mole fraction in the liquid, dimensionless
Y mole fraction in the vapour, dimensionless
z overall mole fraction of the feed, dimension-
less
Greek letters
~t relative volatility, dimensionless
e molar holdup ratio ( = n"/n'), dimensionless
Subscripts
B bottoms
D distillate
F feed
0 at position z = 0
Superscripts
' liquid phase
" vapour phase
* phase equilibrium
REFERENCES
Balakotaiah, V. and Luss, D., 1984, Global analysis of the
multiplicity features of multi-reaction lumped-parameter
systems. Chem. Engng Sci. 39, 865-885.
Bekiaris, N., Meski, G. A. and Morari, M., 1993a, Multiple
steady states in heterogeneous azeotropic distillation.
Technical Memorandum No. CIT-CDS 94-012, California
Institute of Technology, Pasadena, CA 91125, U.S.A.
Bekiaris, N., Meski, G. A., Radu, C. M. and Morari, M.,
1993b, Multiple steady states in homogeneous azeotropic
distillation. Ind. Engng Chem. Res. 32, 2023-2038.
Cairns, B. P. and Furzer, I. A., 1990, Multicomponent three-
phase azeotropic distillation--3. Modern thermodynamic
models and multiple solutions. Ind. Engng Chem. Res. 29,
1383-1395.
Doherty, M. F. and Perkins, J., 1982, On the dynamics of
distillation processes --IV. Uniqueness and stability of the
steady state in homogeneous continuous distillation.
Chem. Engng Sci., 37, 381-392.
Golubitsky, M. and Schaeffer, D. G., 1985, Singularities and
Groups in Bifurcation Theory, Vol. 1. Springer, New York.
 Multiple steady states in binary distillation
Hwang, Y.-L., 1991, Nonlinear wave theory for dynamics of
binary distillation columns. A.1.Ch.E.J. 37, 705-723.
Jacobsen, E. W. and Skogestad, S., 1991a, Control of un-
stable distillation columns. Presented at ACC'91, Boston.
Jacobsen, E. W. and Skogestad, S., 1991b, Multiple steady
states in ideal two-product distillation. A.I.Ch.E.J. 37,
499-511,
Kienle, A., Lauschke, G., Gehrke, V. and Gilles, E. D., 1994,
On the dynamics of the circulation loop reactor--Numer-
ical methods and analysis. Chem. Engn9 Sci. 50,
2361-2375.
Kienle, A., Marquardt, W. and Gilles, E. D., 1992, Steady
state multiplicities in homogeneous azeotropic distillation
processes. Paper 143a, AIChE Ann. Meeting, 1-6 Novem-
ber 1992.
Kienle, A., Marquardt, W. and Gilles, E. D., 1993, Comput-
ing multiple steady states in homogeneous azeotropie dis-
tillation processes, Comput. Chem. Engn9 15 (Suppl.),
37-41.
Laroche, L., Bekiaris, N., Andersen, H. W. and Morari, M.,
1992, Homogeneous azeotropic distillation: separability
and flowsheet synthesis. Ind. Engng Chem. Res. 31,
2190-2209.
2703
Magnussen, T., Michelsen, M. and Fredenslund, A., 1979,
Azeotropic distillation using UNIFAC. Int. Chem. k;ngng
Syrup. Set. 56, 4.2/1-4.2/19.
Marquardt, W., 1988, Nichtlineare Wellenausbreitung ein
Weg zu reduzierten Modellen yon Stofftrennprozessen.
Ph.D. thesis, Universit~it Stuttgart, Dfisseldorf, 1988
Marquardt, W., 1990, Traveling waves in chemical processes.
Int. Chem. Engn9 30, 585-606.
Sridhar, L. N. and Lucia, A., 1989, Analysis and algorithms
for multistage separation processes. Ind. En.qn.q Chem.
Res. 28, 793-803.
Sridhar, L. N. and Lucia, A., 1990, Analysis of multicompo-
nent, multistage separation processes: Fixed temperature
and pressure profiles. Ind. Engn.q Chem. Res. 29.
1668-1675.
Van Dongen, D. B., Doherty, M. F. and Haight, J. R., 1983.
Material stability of multicomponent mixtures and the
multiplicity of solutions to phase-equilibrium equations.
Nonreacting mixtures. Incl. Engn9 Chem. Fundam. 22,
472-485.