Copyright © IFAC Dynamics and Control of Chemical Reactors (DYCORD+'92), Maryland, USA, 1992
DYNAMICS AND CONTROL OF UNSTABLE DISTILLATION COLUMNS E.W. Jacobsen and S. Skogestad Chemical Engineering, University o!Trondheim  NTH, Norway
Abstract
The paper addresses dynamics and control of distillation columns operated at openloop unstable operating points. The fact that ideal twoproduct distillation columns may have multiple steady states and right half plane poles has only recently been recognized . This paper discusses the dynamics and control implications. It is shown that with reflux and boilup as independent variables the operating points become unstable if the internal flows are sufficiently large . An openloop unstable operating point may be stabilized by use of onepoint control , i.e., feedback control of a column composition or temperature. If the control is not sufficiently tight, the column may go into a stable limit cycle. Finally, it is shown that with distillate flow and boilup as independent variables the operating points may become unstable if the level control is not sufficiently tight. The column may also in this case go into a stable limit cycle.
1
a nd instability is most likely to occur [3J. At the end of the paper we consider the DwVconfiguration, and show that in this case there may exist solutions corresponding to stable limit cycles. We will in this paper only consider the multiplicity and instability caused by singularity in the input transformation . The multiplicity and instability that may be caused by including the energybalance will have similar implications for the dynamics and control of di sti llation columns.
2
Introduction
Distillation is undoubtly the most studied unit operation in the process control literature. However , in all previous studies the column dynamics have been assumed to be asymptotically stable (with level and pressure loops closed). The main rea son is that most authors have considered dynamic models with constant molar flows (neglected energy balance) and in addition assumed the inputs (e.g., reflux and boilup) to be given on a molar basis. For the case of molar inputs there exists several papers on uniqueness and asymptotic stability of the operating points in homogeneous distillation. Most papers treat the constant molar flow case (neglected energy balance) , e,g., [5J , [lJ, [6J , [7J , Dohertyand Perkins [2J provide a review of results published in this area, and conclude that, for constant molar flows, multiplicity and instability is impossible for singlestaged "columns" and any multistage column separating a binary mixture. Sridhar and Lucia [8J includ e the energy balance in the model and conclude under certain assumptions that also in this general case binary di sti llation columns will exhibit unique and stable solutions. They do , however , only study a limited set of specifications, namely QDQ8 and LB. However, in a recent paper J acobsen and Skogestad [3J report two kinds of multiplicity which may occur in distillation. 1) Jacobsen and Skogestad argue that columns under operation only in rare cases have all the manipulated inputs on a molar basis. For instance, fixing the valve position will normally correspond closely to fixing the geometric average of mass and volumetric flowrat e. As they show, the transformation from mass or volume flows to molar flows is nonlinear due to the composition dependence and may in some cases become singular, even for the binary case with constant molar flows. A si ngularity in the input transformation will imply that there exist multiple solutions in terms of outputs (compositions) for a given set of inputs (flows). One of the solutions will be unstable . 2) In addition, Jacobsen and Skogestad show that when the energy balance is included in the model. even molar inputs may yield multiple solutions. Both types of multiplicity and instability may be experienced in industrial columns operated with inputs on a mass or volume basis. Jacobsen and Skogestad [3J treat the multiplicity from a steadystate point of view only. In this paper we study the dynami cs and control of columns with multiple solutions. We limit ourselves to discuss mainly one control configuration (set of specifications), namely the case where mass reflux Lw and molar boilup V are used as independent variables. This is the most widespread configuration in industry, and is the configuration for which multiplicity
201
Results on SteadyState Multiplicity in Ideal Distillation
We give here a brief review of the res ults on multiplicity caused by singularity in the input transformation presented in [3J. Consider the twoproduct distillation column in Fig.1. If the feed to the column is given there are at least four flows that may be specified: reflux L, boilup V, distillate D and bottoms flow B. However , for a given column there are only two degrees of freedom at steadystate, that is, only two of these flows may be specified independently. A specific choice of two independent variables is denoted a "configuration" . J acobsen and Skogestad [3J provide an example of steadystate multiplicity in a column separating a mixture of methanol and npropanol. The column has mass reflux and molar boilup as independent variables, i.e., Lw Vconfiguration. Data for the column are given in Table I. Note that the energy balance is excluded , i.e., constant molar flows are assumed. Some steadystate solutions are given in Table 2, and we see that for a specification of mass reflux Lw = 50.0 kg/ min and molar boilup V = 2.0 kmol/min we have three possible solutions I I , I I I and IV in terms of compositions. The multiplicity is graphically illustrated in Fig.2. The observed multiplicity is caused by the transformation between the actual flowrates (mass) and the molar flowrates which determine separation. For a binary mixture the transformation between mass reflux, Lw, and molar reflux , L, is given by
L
= Lw /M;
M
= YDM 1 + (1 YD)M 2
(1)
Here Mi denotes the molecular weight of the individual compo
rf Qo
~~o
'_ _ ~,XB
Figure 1. Two product di stillation column.
IV
Table 2. Steadystate solutions for methanolpropanol column with V =2.0 kmol/min and Lw in the range 48 to 53 kg/min.
V
(l.OO +..:.;;O;.1~ 3G;:....~0::.3:::2:::7:1
!II 0.9
0.0805
YD 0.8
I Il HI IV V
0.00 0.7
II
0.6
3.1
L
D
Lw
kmol/min
kmol/min
kg/min
1.064 1.143 1.463 1.555 1.650
0.936 0.857 0 ..537 0..1 ·1.5 0.350
48.00 50.00 50.00 50.00 53.00
YD
XB
0.534 0.584 0.9237 0.9969
3.10e3 3.50e  3 7.S0e  3 0.101 0.233
0.998~
Conditions for instability
~
1.4
]
1.3
Onestage column Consider the simple column in Fig.3 with one theoretical stage (the reboiler) and a total condenser. Of course, such a column will never be operated in practice because the reflux is simply wasting energy and has no effect on separation. However, we analyze this column due to the simplicity of the dynamic model. As shown in [3], even such a simple column with ideal thermodynamics may have multiple steadystate solution s. Assume binary separation, liquid feed, constant holdup in the reboiler (ML) and negligible holdup in the condenser. The dynamic model of the column becomes:
1.2
dXB AhTt = FZF  DYD  BXB
0 .5c:...~.:.::...:..=~~~~~'
1.7 1.6
:s2
1.5
''I
We have D = V  Land D independent variables we get 48
50
49
51
52
53
54
55
dXB MLTt = F(ZF  XB)
Lu. [kg/min] Figure 2 . Methanolpropanol column: Multiple steady states for LwVconfiguration. Mass reflux Lw is varied while molar boilup V is fixed at 2.0 kmol/min. On the upper plot the
= F and with Land \.' as
+ L(YD 
XB)
+ V(XB
 YD)
SMLc"XB(S)
nents. One might expect the molar reflux to increase monotonically with the mass reflux, that is, (fJL/fJLw)v > O. However , because M is a function of composition, YD, and thereby of Lw, this might not be the case. Assuming molar boilup V fixed and differentiating Lw = LM on both sides with respect to L yields
= Dc"YD(S) 
Bc"XB(S)
Assuming constant relative volatility tion between c"yv(s) and c"XB(S)
c"YD(S) c"XB(S) = (1
0
+ (0 _
 xB)c"L(s)
_
I, (x ) B
;a=h'D+B
1
c"L=Mc"L w +L
ZF
F
0
N
NF
MI
Al2
3.55
8
4
32.0
60.1
(7)
M2  All . AI !,c" XB
(S)
Substituting (8) into (7) we obtain the following transferfunction between liquid composition, c"XB(S), and mass reflux c"L u (s):
(9)
The maximum eigenvalue in selected operating points of the methanolpropanol column with the Lw Vconfigu ration are shown in Fig.2. From the figure we observe that the eigenvalues at the upper and lower branches are negative, implying stability, while those at the intermediate branch (negative slope) are positive. implying in stability of the operating points. Note that the unstable operating points only have a single eigenvalue in the right half plane. The eigenvalues at the singul ar points are zerO as expected.
0.50
(6)
As a is always positive, the pole a/AIL is always negative , implying that all operati ng points are stable when molar reflux Land molar boilup V are ind ependent variables. Now consider mass reflux Lw as an input in stead of molar reflux L = Lw/M. By linearization we obtain for binary separations
OpenLoop Dynamics and Instability for L wV configuration
~IethanolPropanol
(5)
yields the following rela
l)XB)2 
c"XB(S)=YDXBc"L(s) Ahs+ a
For MI < M 2, which is usually the case (the most volatile component has the smallest molecular weight), the second term on the right hand side of (2) will be negative and the total differential may take either sign. The transformation from Lw to L will be singular when (fJ Lw/ fJ L)v = O. A singular point will correspond to a bifurcation point, and the number of solutions changes from one to three. Jacobsen and Skogestad [3] state that solutions with (fJL w/fJL)v < 0 (middle branch in Fig.2) correspond to unstable operating points, but they do not prove this rigorously.
0
+ (YD
Equation (5) then becomes
(2)
Table 1. Data for
(4)
Linearization, Laplace transformation and introduction of dev iation variables assuming F, ZF and V constant yields
corresponding maximum eigenvalue is shown at some of the steadystate solutions.
3
+B
(3)
D, Yo

~ V
L
Column.
F, ZF
Constant molar flows Feed is saturated liquid Total condenser with satu rated reflux Liquid holdups are Ahi/ F = 0.5 min, including reboi ler and condenser.
202
B,XB
~ / Figure 3. denser.
Onestage distillation column with total con
where aw=fI'D+B(YDTB)
11[2 
!If
Ml
I,L
I:~
(10)
The operating point is unstable for a w < O. At steadystate (6) and (7) yield
EJYD) YD  TB ( 7iL \. = D + B / K
(11)
o
and from (2) we find that instability (a w < 0) is equivalent to a negative slope for (dLw/dL)v. This result is in accord a nce with numerical results. Multistage column Jacobse n and Skogestad [4] provide ev idence for the instability of multistage columns _ Using certain assumptions they show that the maximum eigenvalue '\;;,';;~ of a column operating with the Lw Vconfiguration is given by
50
100
150
200
250
300
0 .95 0 .9
YD 0 .85 0.8 0.75
( 12)
0 .7 0.65
Here !If is as defined previou sly and '\~~T is th e maximum eigenvalue for the LVconfiguration. Jacobsen an d Skogestad [4] show th a t instability, ,\~:;~ > 0, corresponds to having a negat ive slope between mass and molar reflux. For details we refer to [~].
3_2
0 .6 0.550!:.5"'0=10*'0''1*50,;;2*00""'2'*5.,,.0,3,.;0"'0'
Effect of operating conditions on stability
J acobsen and Skogestad [3] provide analytical results on when a negative slope between m ass and molar reflux , i.e., in stabili ty, is most likely. They show that a nega tive slope is most likely with large internal flow s (i.e., large L and V) and with intermediat e purities in the top (i.e., intermediate L for given V). This corresponds to having L a nd (EJYD/EJL)v la rge, and according to ( 12 ) this is the case for which in stability is most likely. On thi s basis Jacobsen a nd Skogestad [3] show that there may exist three possible regimes of operation for distillation columns depending on the size of the internal flows: • Intern al flow s low: Unique stable opera ting points. • Internal flow s intermediate: Multipl e steady states, one of whi ch is un stilble. • Internal flow s high: Unique un sta ble operating points.
Effect of Multiplicity and Instability on Column Operation
It is obviou s that the multiplicity a nd instability presented here may have implication s for the operation of di st ill at ion columns using the LwV configuration. In ma nu al (openloop) operation the multiplicity may be experi enced as small changes in the manipulated flows may cause sudd en jumps in the product compositions. Furthermore, one will experience inability to reach a des ired prod uct composition if the separation corresponds to an un st a bl e operating point. However, we show that an unstable opera ting point usually may be easily st abili zed by feed back control.
4.1
singular point in Fig.2 , a nd th e colu m n goes trough a n in sta bility known as a "catastrophi c jump". The ope rator o bserves that the to p composition becomes too pure and th e bottom composition too unpure, and reduces the reflux back to 52.0 kg/min. Howeve r, as seen from Fig.4 this does not have the desired effect, and the operator decides to reduce the reflux all the way back to 50.0 kg/ min. However, due to t he multiplicity th e column now settles in operating point IV with YD = 0.997, that is, hysteresis is experienced in the operation. The simulation s in Fig.4 demonstrate three different effects that may observed in columns wit h multiple steady states: I ) in a bility to reach openloop un stable operating points by manual operation, 2) catastrophic jumps as the column goes through a singular point a nd 3) hy steresis in operation.
4.2
Note that the analytical treatment in [3] was based on id eal separation with constant relative volatility and constant molar flow s.
4
timelminl Figure 4. Nonlinear openloop response of met hanol propanol column to changes in mass reflux Lw. Boilup \ . =2.0 km ol/ min.
Manual operation
We will again consider th e methanolpropanol column in Table l. Assume that th e column initially operates with reflux Lw fi xed at 50.0 kg/m in and boilup V fixed at 2.0 kmol/min. For th ese "alues of Lw and V there exist three possible solution s in term s of compositions accordin g to Table 2 and Fig.2. Onl y two of these a re stable , namely operating points II and IV . Assume t hat the column just has been started up a nd that the product compositi ons corresponds to solution 11 , i. e., YD = 0.584 and TB = 0.00 3.5. whil e the desired operating point is operating point I I I in Table 2, i.e .. YD = 0.9 2~ . In order to in crease th e purity in the top the operator starts to in crease reflux Lw in a stepwi se fashion. Th e reflux is in creased with 0.5 kg/ min every ~o minutes. This is illu st rated in Fig.4 toget her with the response in the top composition. In th e beginning the top composition YD increases slightly with reflux as ex pected. However, as the o perator increases the reflu x from .52 to 52_5 kg/ min the top composition st a rts to in crease drastically. The reason is that the reflu x has been increased passed the lower
203
Operation with composition control
As we have seen, columns operating with mass or volum e inputs may be openloop un stable, and will require feedback co ntrol (in a ddition to level and press ure control) for stabili zation. From control theory it is well known that un stable pol es by them selves do not represent any bandwidth limitations; on the contrary t hey put a lower limit on allowable ba ndwidth of the closed loop system. Problems will therefore on ly arise if there a re bandwidth limitations like right half pl ane zeros at frequ encies comparable to the right half plane pole ("The system goes un stable before we are able to observe what is happening") or if there a re constraints ("we can not counteract the in stability" ). For plants with a RIlP pole p and a RHP zero z one must in general require p < z in order to be able to stabilize the column . With a limited st ru cture of the controller, e.g., a PI controlle r, t he dist a nce between p and z mu st be larger. Good control of disti llati on column s usually requires twopoint cont rol , i.e., feedback control of both product compositions. However, in order to stabilize an openloop un st abl e colum n onepoint co ntrol will suffice. This is also the way most industrial column s with composition (Ontrol are opera ted. An un stable column operat in g with the Lu.Fconfiguration may be stabili zed by control lin g either top or bottom composit ion, or any oth er variable related to composition . c.g. a temperat ure in side the column. For operati ng point III of the methanolpropanol column th e RHP pole is at p = O.08Gmill 1 and we are un able to st ab ili ze th e column with a PI controller when the deadtime exceeds 11 min. (z '" 2/()d = 0.182min 1 ). Howeve r, composi ti on measurements in indu strial columns (GCanaly sis) may typically have deadtimes up to 30 min. , a nd one shou ld then use faster temperature measureme nt s in order to stabili ze the colum n. Nonlinear Simulations. Fig.5 shows nonlinea r simulati ons of th e methanolpropanol column using a singlel oop PI controll er 1 between top com posi ti on YD an d mass refl ux Lu' with a 1 minute measu remen t dead time illclud cd. 1.!olar boilup V is kept con
stant at 2.0 kmol/min. The figure shows the responses to set point changes in YD from operating point 11 (openloop stable) to operating point III (openloop unstable) and then further on to operating point IV (openloop stable). A logarithmic measurement YD = 1nl 1 YD) was used in the controller as this reduces the nonlinearity of the initial response between different operating points [9J. From the figure we see that the controller is able to stabilize the openloop unstable operating point III with a RHP pole at 0.086 minI. The simulations also show that the same controller may be used in these three widely differing operating points. The reason is that the initial response (highfrequency dynamics) in terms of logarithmic composition YD is similar in all operating points. We would get instability if we used mole fractions, YD . as is done conventionally. From the plot of massreflux against time we see that the steadystate change in the input is zero. showing that the three operating points are multiple solutions. One should be careful about detuning a controller in an openloop unstable process as the bandwidth may become lower than the minimum allowable and the operating point becomes closedloop unstable. This is illustrated in Fig.6. where the controller gain has been reduced by a factor of two compared to the simulations in Fig.5. Operating point II I is now closedloop unstable , and a small setpoint change makes the system start drifting away. However, this does not imply that the column goes globally unstable in the sense that physical constraints are violated. Since there exists steadystate solutions above and below the unstable solution the column goes into a stable limit cycle. If the controll er gain is reduced further the limit cycle will continue, but now with a longer period of each cycle and with higher peaks in composition. There will also exist cases where there are no solutions either abo\'(, or below the unstable solution. In this case the column is likely to go globally unstable as either the condenser (missing upper branch) or reboiler (missing lower branch) would run dry. For a discussion of the implications of open loop instability for twopoint control we refer to [.IJ.
5
Other Bifurcation Parameters
So far we have only considered the inputs, e.g. reflux L ll • and boilup l', as potential bifurcation parameters. That is, in all studies we have assumed the other parameters, e.g., feed flow F , feed composition ZF, feed liquid fraction qF, tray efficiency etc .. to be fixed. However , it is clear that these parameters will vary during operation and may. similarly to the inputs , cause the column operation to go from openloop stable to openloop unstable. To illustrate this consider Fig.7 which shows steadystate solutions for the methanolpropanol column with Lw = 50.0 kg/min. V = 2.0 kmol/min and feed composition ZF in the range 0.10 to 0.60. From the figure we see that there are multiple solutions for ZF in the range 0..16 to 0.54. This implies that disturbances in the feed composition may cause the column to go through a singular point and thereby "jump" to another solution branch. This is illustrated in Figure 8 which shows the response in top composition YD to a change in feed composition ZF from 0.50 (operating point IV in Table 2) to 0.16. The figure illustrates how the top composition "jumps" to the lower solution branch and settles in operating point V I. When the feed composition returns to ZF = 0.50 the solution remains on the lower branch and settles in operating point
II.
Instability with the Dw V Configuration
6
"Ve have so far only considered using reflux and boilup as independent variables , i.e. , the L,,'vconfiguration. This is also the 1lI0st widespread configuration in industry. However, there are many different configurations that may be used. For instance, changing condenser level control from using distillate Dw to using reflux Lw results in the Dw V configuration. For all the examples we have have studied this configuration yields a unique steadystate solution in terms of compositions. Furthermore, we have assumed perfect level control, in which the operating point is found to be
ITuned to yield reasonably fast response. Note that Ziegler.Nichols tuning rules resulted in a closed.loop unslable system. 1.05'r~~~~~~1
0.95
IV
III 0.9
0.95
0.85
0.9
Yv
0.85
0.8
Yv 0.8 0.75
0.75
0.7
0.7
0.65 0.6 0.55
o 650!c,,""'0"02'"0'"'0;;;3ctiOnO4MOirO'"51iOnO6MOO
r1L 0
50
100
150
200
250
time [minJ
350
300
62
 cc;::::=~=:::::::::___=
60 58
C
'E
=. bD
...:::,'
yv
56 54 52
IV
50 II
0.6 48 46
0
50
100
150
200
250
300
0~L,;.4~8~4~9~50n5~',,5~2,,5~3,.5~4~~55 Lu' [kg/minJ
350
time [111il1J Figure 5. Nonlinear simulation of methanol propanol column with onepoint control of topcomposition YD using mass reflux Lw. Setpoint changes from operating point I I to II I and from III to IV. lloilup V = 2.0 kmol / min. Controller parameters: le = 3.0 and TJ = 11.0 min. Gain is for logarithmic composition, i.e., log( 1  YD).
Figure 6. ;\onlinear simulation of methanol propanol column with onepoint control of topcomposition YD using mass refl ux L ll" Controller gain reduced by a factor of 2 COIllpared to Fig ..,>. Upper plot: Time as independent variable. Lower plot: Phaseplane plot. Dashed line shows steadystate solutions.
204
6.1
IV 0.9
To understand why the steadystate for the Dw Vconfiguration becomes unstable, consider the transfer function (8YD/DD w )v (8) which may be written
III
Yv
Analytical treatment
0.8
8 YD ) (8)= (8 YD ) (8)(8Lw) (8) ( 8Dw v 8Lw v 8Dw v
0.7
Here the transfer function (8YD/8L w)v (8) expresses the effect of reflux on top composition with the Lw V configuration, and we have seen that it may be unstable with a single RHP pole. For simplicity we consider only the largest pole in the transfer function
II
0.6 0.5
0'~4
0.42 044 046 0.48
0.5
(13)
0.52 0.54 0.56 0.58
8 yD ) (s) = _ k ( 8Lw v 8  a
0.6
z['
(14)
Here a denotes the maximum eigenvalue for the Lw Vconfiguration. Th(' transfer function (8Lw/8Dw)v (s) may be computed from a material balance around the condenser
Figure 7 Steadystate solutions as a function of feed composition ZF for methanolpropanol column. Reflux Lu. = 50 kg/ min , Boilup V = 2.0 kmol / min.
(15)
IV Differentiation of ( 15) yields 0.9 0.8
8Lw) (8) =_[.;MD (VT(M 1  M 2) (88 YT ) ( 8Dw Dw v AMD + s
YD
He re YT denotes the composition of VT. We assume negligible condenser holdup so that (8yT/8Dw)v (s) = (8YD/8D w)v (8). Inserting (14) and (16) into (13) yields
0.7
1I
0.6
0.5
(8) 1) (16) V
8YD ) ( 8Dw v (s) =
l ___ ~_~ ________________________~~_~ _____ j 
82
+ (I\MD
I\MDk  a)s  I\MD(a
+ kVT(M 1 
M 2 )) (17)
The poles of the transfer function (17) become
0.401;5;t;0,,~10t;;0,,.1c:5"0"'20~0~,2"'5"O""300
timefminl
'xl.2
Figure 8 Nonlinear openloop simulation of methanolpropanol column for changes in ZF. Reflux Lw = 50 kg/ min and boilup V = 2.0 kmol/min. Roman numbers I I  IV refer to Table 2 with ZF = 0.50. Operating point V I corresponds to ZF = 0.46. asymptotically stable. However, here we show that without the assumption of perfect level control the operating point may become unstable also with the Dw Vconfiguration. We start by considering an example and will then explain the results thereof using analytical results.
=
~([(MDa)±~J(I\MD  aF + 4[(MD(a + kVT(M 1 
M2»
(18) Let us now use (18) to consider the stability of the Dw Vconfiguration for the two cases when the pole a of the Lw V2
0.924r~~~~.~~~~~~,
YD
a)
O 924 .
0.923 0.923 0.923
Example. We will again consider the methanolpropanol column in Table 1. The holdups in the reboiler and condenser are in creased to MD/ F = MB / F= 5.0 min. We consider the case with constant molar flows, and use distillate flow Dw and boil up V as ind ependent inputs , i.e., Dw Vconfiguration . With this configuration the condenser level is controlled by reflux Lw and the reboiler level is controlled by bottoms flow Bw. The nominal operating point we consider has Dw= 18.36 kg/min and V =2.0 kmol/min. For these specifications we obtain YD=0.9237 and xB=0.0078 , and the steadystate is unique. Note that the operating point correspond s to solution III for the LwVconfiguration in Table 2 and Fig.2. that is , the operating point is unstable with reflux and boilup as independent variabl es. We now consider the stability of th e operating point for different gains [(MD in the condenser level controller. A pure proportional controller is used , i.e., dL w(8) = I\M DdMD w(8). We assume perfect level control in the reboiler. Fig.9a shows the respon se in top composition to a small increase in D w , keeping V constant , with level control gain [{MD=0.05. We see that the respon se is oscillatory but asymptotically stable, i.e. , a stable spiral. Fig.9b shows the corresponding response with [(MD reduced to 0.03 , and we see that the steady state now is an unstable spiral. However, the response sett les into a stable periodic solution, that is. a stable limit cycle. As the steadystate changes from a stable spiral to an un stable spiral as the level control gain is reduced impli es that a pair of complex conjugate eigenvalues cross the imaginary axi s. The fact that a stable limit cycle appears as the steadystate becomes unstable, implies that the system goes through a dynamic bifurca tion known as the Hopf bifurcation.
205
0.923 0.923 0.922 0.922 0.9220
200
400
600
800 1000 1200 1400 1600 1800 2000
b) A 095~/\1
YD
v
09
r
fI
f\
f\
f\
V
0.85
0.8
0.750
200
400
600
800
1000
1200
1400
1600
time [min[
Figure 9 Nonlinear openloop simulation of methanolpropanol column with DwVconfiguration . Responses to small increase in Dw with different gains [(MD in condenser level controller. a) [(M D=0.05 . b) [{MD =0 .03. Boilup V=2.0 kmol / min.
References
configura tion is in the LIIP and RIIP, res pectively : I) Stable LwVconfiguration . a < 0: In thi s case the first t erm in (1 8 ) is negative for all values of 11"\/0 > O. Furthermore. th e second t erm under the root in (1 8 ) is nega tive and th e root will be real with a value less th a n (I'M o  a) or it will be imagin a ry. Thi s implies th a t both eige nvalues in ( 18 ) are in the LHP, th a t is. the D wV confi guration is st a ble fo r all values of 1I'Mo > O. 2) Unst a ble Lwy' .config urati o n, a > 0: In thi s case we have that the first t erm in ( 18) is positive if 11'.1/ 0 < a, that is. at least one of the eigenvalu es in ( 18) are in the RHP with 1I'Mo < a. Th e size of 1I'Mo will determin e whether the root in ( 18) is imaginary. For 1\.\10 = a , i.e. , the bifurca tion po int , we have that the root is im aginary if kVT ( M I  M 2 ) < a. whi ch is the case in all exam ples we have studi ed. We conclude from the above analys is that a prerequi sit e for instability with the DwV·configurati on is that the operating point is unstable with the LwV.configuratio n. This is not too surpri sing as the level control for the DwVconfiguration may be viewed as a feedback effect on the LwV configura tio n. If the feedback control is not tight eno ugh , we a re not a ble t o st a bili ze the column , whi ch is similar to what we found for t he case of one poi nt control wit h the LwVconfiguration. \\,ith a gain I'Mo =O , i.e., no condenser level control, we see from ( 18) tha t t here will be a RHP pole at a ( in addition to a pole at 0) , and We effec tively have the stability properti es of the Lw V ·configuration . In our example we find th at th e LwV. configuration is unstable wi t h a pole a = O.OH (with MD / F = M8 / F = 5min. ) and from (1 8) we predict in stability for th e D wVconfi guration with 1I'Mo < 0.041 . Fro m the full model we find th a t in st ability o cc urs for I'Mo < 0.043. Th e deviation in predi cted and computed val ue is ex plain ed by our assumpti o ns of fI rst ·order res po nse in ( 14) a nd negli gible cond enser hold up in t he a nalyti ca l t rea t ment.
7
[I] Acrivos, A. and N.R. Amund son, 1955, "Appli cation of Matrix Mathematics to Chemical Engineering Problems", Ind.Eng.Chem ., 47,8,15331541 [2] Do herty, M.F . and J.D . Perkins , 1982 , " On the Dyn a mics of Di stillation ProcessesIV. Uniqueness and Stability of the Steady State in Homogeneous Continuous Di still ation " , Chem.Eng.Sci, 37, 3, 38 1392 [3] Jacobsen, E.W . and S. Skoges tad, 1991a, " Multiple Stead y States in Ideal Twoproduct Di stillation" , A lChE J ., 37, 4, 499511. [4] Jacobsen, E .W. and S. Skoges tad, 1991b , "Control of Unstable Distillation Column s", Proc. of 1991 American Control
Conference. [5] Lapidu s, 1. and N.R. Amund son , 1950 , "Stagewi se Absorpti on and Extraction Equipment  Transient a nd Unsteady St a te Operation " , Ind.Eng.C hem., 42 , 1071  1078. [6] Rosenbrock, H.H., 1960 , "A Theorem of " Dy namic Co nserva ti on" for Di stillation " , Tran s. lnstn. Chem.Engrs., 38 , 20, 279287. [7] Rose nbrock , H.H., 1962 , "A Lya punov Function with Applicatio ns to So me Nonlinear Physical Problems", Automatica, 1,31 53. [8] Sridhar , L.N. and A. Lucia, 1989, " Analysis and Algorithms for Multi stage Separa ti on Processes" , l f3 EC Res., 28. 79 3803. [9] Skogest ad , S. and M. Morari , 1988, "U nderstanding th e Dynam ic Behavior of Di still ati on Column s", lnd. f3 Eng. Chem. Research, 27, 10, 1848 1862 .
Conclusions
• Di stilla ti on columns operatin g with the LwV·confi g ura ti o n may ha.ve m ultiple steady st ates a.n d un stable operatin g po in ts . Th e probability of in st a bility in creases with in creased int ern al fl ows. • Th e multipli city and inst a bility may cause pro blem s for m anu al opera tion of di stillation co lumn s. Three effec t s may be obser ved: I ) catastrophi c jump ph enomen a , 2) in a bility to reac h cert a in sep· a ra tion s a nd 3) hysteresis in o pera ti o n. • An openloop un st a bl e o pera ti ng poin t may usuall y be easil y st ab ili zed by feedback co nt rol of a co mpositi o n o r t empera ture. If t he ba nd width is not suffi cient ly tigli t the column may go into a sta bl e limit cycle. • Col um ns opera tin g with th e D wl' ·confi g ura ti o n may beco me un st abl e if the le"el co nt rol is no t suffi cient ly ti g ht. Th e in sta bility will co rres po nd to a lI o pf bifurca ti o n. a nd a sta ble limi t cy cl e ap pears. NOMENCLATURE (see a lso Fig .l )
B  botto ms flow (kmol / min) D  di still ate fiow (kmol / min ) F  feed ra te (kmol / min) L  reflu x fl ow rat e ( kmol/ min ) M  mole weight . usuall y of to p prod uct ( kg / kmol ) JII J  pure component mole weight of most vola t ile co m ponent ( kg/ kmol ) ,\[2  pure com ponent mole weight of leas t vol at il e component ( kg / kmol ) N  no. of theoretic al st ages in colu mn SF  feed st age loc ati o n ( Ireboi ler ) \' _ boi lup (km ol/ mi n) (dete rmin ed indi rec tly by hea tin g) XB  mole fraction of mos t volatil e co mp onent in bott o m prod uct YD . mole fr ac ti o n of mos t vola til e com po nent in to p produ ct ZF  mole fr ac ti o n of most vo la tile co mponent in feed GrEek symbols Cl
=
y'v' )' rd at ive volat ili tv (b in a ry mixt ure)
( 1 y,) ( 1
I,
•
Ama, = maXi IA,( ..I )1 · maximu m eige nvalu e = do min a nt pole
S lIbsc ripts w  fl ow ra te in kg/ min
206