Sei sulla pagina 1di 50

OPERABILITY OF CHEMICAL REACTORS: MULTIPLICITY BEHAVIOR OF A JACKETED STYRENE POLYMERIZATION REACTOR

Louis P. Russo and B. Wayne Bequette

Howard P. Isermann Department of Chemical Engineering Rensselaer Polytechnic Institute Troy, NY 12180-3590 bequeb@rpi.edu Fax: (518)276-4030 Tel: (518)276-6683 Submitted to Chemical Engineering Science July 1996

Keywords.

Process Design, Operability, Steady-state Multiplicity, Polymerization, Bifurcation.

ABSTRACT
Steady-state multiplicity analysis is used to study the operability of a jacketed exothermic styrene polymerization CSTR. The global multiplicity behavior is characterized in terms of an initiator decomposition Damkohler number and the dimensionless heat of reaction. Low and high temperature infeasibility regions are shown to occur under certain design conditions. The approach developed is based on using a combination of analytical and numerical techniques to guide the analysis, rather than using a purely numerical approach. A reactor design used by previous researchers to show poor closed-loop performance from a single input-single output control strategy is studied in this paper. It is shown how this poor performance could have been predicted, and eliminated, at the process design stage using our bifurcation-based approach.

1 Introduction
The integration of process design and control (or operability) has been recognized as an important objective for many years. A number of relevant techniques have been proposed, including exibility analysis (based on nonlinear programming and steady-state models), resiliency (based on linear, multivariable models), and steady-state and dynamic \back-o " analysis (where the actual operating point is chosen by \backing o " from the optimum point which lies at the intersection of constraints). In this paper we present an approach, based on bifurcation (singularity) theory, to understand how design parameters in uence the characteristic input/output behavior of process systems. We study the steady-state multiplicity behavior of a jacketed exothermic styrene polymerization CSTR. Since the structure and properties of polymers are determined primarily by the operation of the chemical reactor in which they are made, there is signi cant motivation to optimize the process design and operation to improve controllability. Previous researchers showed poor single input-single output control performance for a given set of polymerization CSTR design parameters. In this work we show how this poor performance could have been predicted, and eliminated, at the process design stage using our bifurcation-based approach. The bifurcation results presented in Russo and Bequette (1995, 1996a, 1996b) are particularly important from a practical \controllability" point of view. This analysis demonstrates that in some regions of the parameter space there does not exist any manipulated variable value which will allow operation at a steady-state setpoint (using the cooling jacket owrate as the single manipulated variable). Russo and Bequette (1995) showed that under certain conditions a two-state CSTR model (neglecting cooling jacket temperature dynamics) could have a monotonic steady-state relationship between the reactor temperature and cooling jacket temperature, while a three-state CSTR model operated under the same conditions exhibited multiple steady-state behavior between the reactor temperature and the cooling jacket owrate. Thus, we see that an analysis of the two-state CSTR model could lead one to believe that the system is inherently \safer" and easier to operate than is the case as predicted by a more realistic three-state CSTR model incorporating a cooling jacket energy balance. We are motivated in our studies to consider these attributes of \process controllability" and \process operability" in order to understand how modeling decisions (and ultimately, design and operation decisions) e ect the operating characteristics of the chemical reactor.

2 Industrially Relevant Application


Although a detailed analysis of the bifurcation behavior of a jacketed exothermic CSTR was presented in Russo and Bequette (1995, 1996a), the kinetic mechanism studied up to now is quite simple in nature and hence is not generally su cient to capture the behavior of some industrially relevant chemistries. Therefore, it is important to examine a chemical \system" in which these bifurcation results can be of practical usage. The production of synthetic polymers is an example of a chemical system that is important from an industrial viewpoint. Synthetic polymer production exceeds one hundred million metric tons per year worldwide (Ray, 1993). Hence, there is su cient incentive to obtain a deeper understanding of the parameters which in uence the steady-state and dynamic behavior of polymerization chemical reactors. There are many di erent types of polymers available and depending on the properties that they exhibit, polymers range from low volume, high value added specialty materials at several thousand dollars per kilogram to high volume material for approximately one dollar per kilogram (Ray, 1993). The structure and properties of the polymer are determined primarily by the operation of the reactor in which they are made. Therefore, it is important to have tight control of the operating conditions inside the process. In order to do this in an e ective manner, one should have a good understanding of the underlying kinetics, thermodynamics, and transport properties associated with the polymerization reactor. It is essential to have reliable and accurate sensors to e ectively monitor the conditions inside the reactor so that polymer with the desired properties is produced. Polymer science is a very broad eld of study, and as such there exists a wealth of literature on the subject. The emphasis in this paper is on an understanding of the bifurcation behavior of exothermic polymerization reactors. Ray (1993) serves as an excellent review of polymerization reaction modeling and control, while Billmeyer (1984) presents a more general overview of polymer science.

3 Review of Previous Bifurcation Results


The open-loop steady-state and dynamic behavior of polymerization reactors has been the focus of much research during the last twenty- ve years. A primary reason behind this research is the tremendous increase in demand for polymeric materials. The purpose of this section is to review some of the more relevant literature on the bifurcation behavior of polymerization reactors. Jaisinghani and Ray (1977) studied the bulk homopolymerizations of methyl methacrylate (MMA) and polystyrene in a CSTR. They determined the in uence of reactor operating conditions on the types of steady-state and dynamic behavior possible. Jaisinghani and Ray (1977) initially studied a simpler model for the reaction kinetics (assuming constant initiator concentration). They found evidence for the presence of steady-state multiplicity and periodic (Hopf bifurcation) behavior, but were only able to nd unstable limit cycles in their analysis. A \gel e ect" which predicted that the termination rate could decrease up to ve orders of magnitude under the right operating conditions was included in the models studied in Jaisinghani and Ray (1977). The \gel e ect" (also known as autoacceleration) is marked by a signi cant increase in reaction rate and molecular weight and results from a decrease in the rate at which the polymer molecule di uses through the reaction solution (Billmeyer, 1984). The termination step is generally di usion controlled for liquid-phase polymerizations, hence a decrease in di usion (due to an increase of viscosity with high conversion) results in a decrease in the termination rate (Billmeyer, 1984). This leads to an increase in overall polymerization rate and molecular weight. Brooks (1981) considered the dynamic behavior of the bulk polymerization of styrene in a CSTR. One important point of this work was that more sophisticated models which did not assume constant heat capacity or constant density in the reactor were employed. It was shown that more realistic models for these parameters were very important during the startup phase of the reactor; the constant parameter assumptions were adequate only for steady-state analysis. Schmidt and Ray (1981) analyzed the multiplicity behavior of MMA solution polymerization in an isothermal reactor operating at the boiling point of the reactor mixture. They found that the multiplicity behavior in this case was a result of the strong \gel e ect" associated with the process. Multiple steady-state behavior occurred as the solvent volume fraction was decreased. Hamer et al. (1981) presented the rst account of stable limit cycles for the polymerization of 5

MMA. They studied the homopolymerizations of MMA and vinyl acetate, and their copolymerization. The steady-state and dynamic behavior was classi ed in the parameter space by varying system parameters and using the Routh-Hurwitz criteria to give conditions on the characteristic equation coe cients for a speci ed eigenvalue structure. Schmidt et al. (1984) demonstrated the experimental existence of multiple steady-states under nonisothermal reactor conditions, and presented experimental veri cation of steady-state isolas for the homopolymerizations of MMA and vinyl acetate. Previous modeling studies were used to guide experiments in nding multiplicity behavior. Choi (1986) compared the bifurcation results for two models of the homopolymerization of MMA. The rst model utilized the \standard" kinetic description (common to free radical initiation), whereas the second model assumed that the initiator concentration was constant. A slightly di erent dimensionless group parameterization was used which provided a more detailed steady-state structure, as compared to previous work. Teymour and Ray (1989) performed a bifurcation analysis of a model of an experimental polymerization reactor. The model parameters used were obtained from the literature (no adjustment was made to these parameters for the present experimental system). They presented the rst experimental evidence of limit cycle behavior in the solution polymerization of vinyl acetate, and classi ed the steady-state and dynamic behavior in the parameter space. This analysis was used to guide experiments to nd regions of process parameters where limit cycles were predicted to occur. Adebekun et al. (1989) analyzed the steady-state multiplicity behavior possible in the polymerization of MMA in a CSTR. The primary focus of this work was to include polymer properties such as the molecular weight distribution (MWD), number average chain length, and polydispersity of the MWD in the bifurcation analysis. They rea rmed work by Schmidt and Ray (1981) which indicated that the multiplicity behavior in an isothermal polymerization reactor was the result of the \gel e ect". The presence of two steady-state isolas over a very small region of solvent volume fractions was demonstrated. Kim et al. (1991) studied the homopolymerization of styrene in a series of two CSTR's. One important point made in this work was that an understanding of the reactor behavior is important for the design of control systems for producing polymers with the desired physical properties. The use of more than one reactor was justi ed by noting that often the rst reactor serves to prepolymerize the monomer. Kim et al. (1991) found limit point bifurcations, Hopf bifurcations, in nite periodic orbits, period doubling bifurcations, isola behavior, and multiple Hopf bifurcation 6

points. Teymour and Ray (1992a) presented an extensive study of limit cycle behavior in an experimental laboratory reactor. They found that a major problem with studying the dynamics of small (lab-scale) reactors is the damping e ect of reactor externals such as the reactor wall, impellor blades and shaft, etc. Teymour and Ray (1992a) found that either increasing the residence time or decreasing the coolant temperature lead to limit cycle behavior in the experimental polymerization reactor. The model was able to predict the oscillation amplitudes and limit cycle periods well, with only small adjustments in a heat transfer coe cient. Teymour and Ray (1992b) focused their analysis on a full-scale polymerization reactor. They modi ed the model used in previous studies to more accurately describe conditions in an industrial polymerization reactor. The bifurcation analysis package AUTO (Doedel, 1986) was used to obtain an understanding of the di erent types of behavior possible in this reactor system. The full-scale model demonstrated even more complex dynamics than the lab-scale model. Pinto and Ray (1995a) constructed a mathematical model of the copolymerization of MMA and vinyl acetate in a lab-scale reactor. The bifurcation behavior of the system was examined using AUTO (Doedel, 1986) and the e ect of perturbations in the heat transfer coe cient and the coolant temperature were studied. Experimental support for limit cycle behavior demonstrated that decreasing the percentage of MMA monomer in the feed removed the stable limit cycle. Pinto and Ray (1995b) studied the copolymerization of MMA and vinyl acetate in a lab-scale reactor. They demonstrated that the system dynamics were very sensitive to small changes in the feed conditions, particularly, the initiator feed concentration. In an extremely small area of parameter space, two isolas were predicted. However, they determined that the appearance of the second isola was related to the \gel e ect" correlation and may not be an actual feature of the physical system. They showed that a boiling constraint that arises due to operation at atmospheric conditions was responsible for regions of up to ve steady-states. Multiple stable periodic orbits and isolas of periodic solutions were also found.

4 Bifurcation Behavior of a Jacketed Exothermic Polymerization Reactor


4.1 Motivation
One can surmise from the review in the previous section that there has been substantial research activity in polymerization reactors. This relatively cursory review focused primarily on the studies concerned with the bifurcation behavior. Overall, the general trend has been from analyzing simpler kinetic models initially to more sophisticated chemistries such as copolymerization of two monomers. A common theme in these studies is that a variety of industrially relevant polymerization reactions are highly exothermic, and thus have interesting steady-state and dynamic behavior which may make these systems di cult to operate and control. Kim et al. (1991) note that poor heat transfer from the reactor to the cooling jacket in polymerization systems is one of the major causes for strong process nonlinearity. A major problem is that industrial scale reactors have heat transfer area per volume ratios which are much smaller than lab-scale units. A fundamental question to consider is thus: How can one e ectively \remove" the energy liberated in highly exothermic reactions such as are encountered in polymerization systems? Perhaps the easiest way to accomplish this is to add a \cold" stream directly to the reaction mixture; this a ords e cient and e ective heat transfer since the entire volume is contacted with the heat transfer medium. A major problem in doing this is that one is diluting the reactor contents, hence a larger residence time is needed. The addition of a \foreign" chemical to the reaction mixture can change the polymer properties. Another alternative is to operate the reactor at the boiling point of the solvent, and to recondense and recirculate the solvent to the reaction mixture. This e ectively limits the reactor to a small range of operating temperatures (near isothermal behavior, as the boiling point of the solvent - monomer - polymer mixture will change as more monomer is converted to polymer). In order to achieve high conversion one may have to pressurize the reactor. However, this may have an adverse e ect on the reaction kinetics. Perhaps the easiest and most common approach is to utilize a cooling jacket to remove the energy liberated by the reaction mixture. A \cold" liquid (water, Dowtherm, glycol, etc.) is pumped through the jacket to remove excess heat from the reactor. The cooling jacket owrate is the 8

primary manipulated variable used to control the reactor temperature (which in turn speci es the conversion and thus the polymer properties). It is worthwhile to note that none of the studies reviewed considered the e ect of a cooling jacket on the steady-state and dynamic behavior of polymerization CSTR's. Congalidis et al. (1989) assumed that a \properly tuned" PI controller could be used to manipulate the coolant ow rate in order to obtain any value of the reactor jacket temperature. However, it was demonstrated in Russo and Bequette (1995) that for even simple 1st order kinetics (A ! B ) that including a thermal energy balance around the cooling jacket lead to interesting bifurcation results which are not present in a lower order (two-state) model of the reactor. Perhaps the most important result obtained was that even for the simple kinetic model used, the presence of infeasible reactor operating regions is predicted. Therefore, it is a poor assumption to neglect the cooling jacket energy balance in the control system design. It is well known that constraints on input variables can result in severe di culties in obtaining robust control. However, these constraints are not due to any process equipment limits (such as a maximum owrate for a pump) but are chosen as physical limits on the cooling jacket owrate. Pinto and Ray (1995b) stressed that the role of physical constraints on the model has been largely overlooked in the past. These constraints, taken with the mathematical model, create a new model which may or may not present completely di erent bifurcation structures that can improve the design and operation of real processes (Pinto and Ray, 1995b). Since the cooling jacket owrate is the primary manipulated variable for control of reactor temperature, it is crucial to use it as the bifurcation parameter in order to understand how it in uences the behavior of the system.

4.2 Review of Hidalgo and Brosilow (1990)


One interesting study which incorporated an energy balance around the cooling jacket was performed by Hidalgo and Brosilow (1990). Hidalgo and Brosilow implemented a low order nonlinear model predictive control (NMPC) algorithm for the control of styrene polymerization in a CSTR. A combination of NMPC and coordinated control provided safe and e ective control about an unstable operating point. Coordinated control consists of using two or more manipulated inputs to maintain a single output at its desired setpoint (multiple input single output or MISO control). 9

In this case, the cooling jacket owrate was the primary manipulated variable while the monomer owrate (the secondary manipulated variable) was only \changed" if the cooling jacket owrate \hit" a constraint. A number of di erent control case studies were conducted by Hidalgo and Brosilow (1990). The response of a conventional SISO controller (manipulating cooling jacket owrate alone) was compared to a MISO controller (coordinated control manipulating cooling jacket owrate and monomer owrate). The performance of the MISO control strategy was substantially better than the SISO control strategy. In fact, for a large enough setpoint change, the SISO control strategy is unstable. However, Hidalgo and Brosilow never explain why the SISO controller performs so poorly, especially in light of the fact that a nonlinear model of the process is used which incorporates a great deal of process information in the control strategy. Hidalgo and Brosilow (1990) note that \safe control of an unstable process requires the availability of a su cient control action to keep the process in a small region about the desired operating point in spite of disturbances and control e ort constraints." They claim that the styrene polymerization CSTR \can not be safely controlled by manipulating only the cooling jacket owrate", yet no physical explanation is presented. It has been surmised from the review section that few if any bifurcation studies have been conducted on a jacketed polymerization CSTR. This fact coupled with the intriguing results of Hidalgo and Brosilow (1990) make this polymerization system an ideal choice for further study.

5 Styrene Polymerization CSTR Model


In this section the features of the styrene polymerization model of Hidalgo and Brosilow (1990) are described. The free radical polymerization kinetic mechanism is modeled as:

10

kd ?!

2R

(initiator decomposition) (chain initiation) (propagation) (termination) (termination)

ki M + R ?! kp Pn + M ?!

P1 Pn+1

ktd T n + Tm Pn + Pm ?! ktc Pn + Pm ?!

Tn+m

where I is the initiator, R is the radical produced by initiator decomposition, M is the monomer species, P is the growing polymer chain, and T is the terminated polymer chain (cf. Billmeyer (1984) for a discussion on free radical polymerization kinetics). A number of assumptions were used by Hidalgo and Brosilow (1990) in obtaining the styrene polymerization CSTR model. In particular, the quasi-steady-state approximation and the long chain hypothesis were employed in the kinetic modeling. The quasi-steady-state approximation can be made by realizing that the lifetimes of the radical species R, P are short compared to the system time constants (Jaisinghani and Ray, 1977). Hence, the net reaction rate of the radicals is near zero (rR 0, rP 0). The long chain hypothesis states that most of the monomer is added to existing polymer chains as opposed to starting new polymer chains. In addition, the \gel e ect" (a decrease in termination rate associated with increased viscosity as the reaction proceeds) was neglected, since the reactor was operated at a relatively high solvent volume fraction. Choi (1986) notes that this is generally a good assumption for high solvent volume fractions. Termination for the styrene polymerization is generally due to combination (i.e. kt = ktc ). Physical properties of the reaction mixture such as density, heat capacity, heat transfer coe cient, etc. were assumed constant. The reader is referred to Appendix 1 and Hidalgo and Brosilow (1990) for additional details concerning the modeling assumptions and derivation of the modeling equations. The model for polystyrene polymerization as given by Hidalgo and Brosilow (1990) is: dCi = (QiCif ? (Qi + Qm + Qs )Ci) ? k C
d i V dCm = (QmCmf ? (Qi + Qm + Qs)Cm) ? k C C p m gp dt V

dt

(1) (2)

11

reactor temperature, cooling jacket temperature, concentration of growing polymer, and cooling jacket owrate, respectively. The initiator is azobisisobutyronitrile (AIBN) dissolved in benzene, while the monomer is styrene and the solvent is benzene. The process ow diagram for the styrene polymerization CSTR with a cooling jacket is shown in Figure 1. Five di erent sets of numerical values of the process parameters are given in Hidalgo and Brosilow (1990). The variables of the styrene polymerization model are de ned in Table 1, while a representative selection of the process parameters is shown in Table 2. No analysis of the multiplicity or open-loop dynamic behavior of this model has been performed. The four-state polymerization CSTR model can be extended in order to calculate the numberaverage molecular weight (NAMW) Mn . Following Schmidt and Ray (1981), two polymer chain moment equations are appended to the current model:

dT = (Qi + Qm + Qs)(Tf ? T ) + (? Hr ) k C C ? UA (T ? T ) (3) c dt V Cp p m gp CpV dTc = Qc(Tcf ? Tc) + UA (T ? T ) (4) c dt Vc c Cpc Vc s d Ci Cgp = 2fk (5) kt where Ci , Cm, T , Tc , Cgp , and Qc represent the concentration of initiator, concentration of monomer,

where 0 is the total molar concentration of dead polymer and 1 is the total molar concentration of monomer units present as polymer (Schmidt and Ray, 1981). It can be shown (Appendix 2) that equation 7 becomes: d 1 = k C C ? Qt (8) p m gp V 1 dt The number-average molecular weight (NAMW) Mn is de ned as:
P1 i=1 Mi Ni P 1

d 0 = 1 k C 2 ? Qt dt 2 tc gp V 0 d 1 = 1 k C 2 ? Qt dt 1 ? tc gp V

(6)
1

(7)

Mn =

i=1 Ni

(9)

where Mi is the molecular weight of the ith polymer and Ni is the number of molecules of type i (Billmeyer, 1984). For most polymers the number-average molecular weight (NAMW) lies close to the peak of the molecular weight distribution curve (Billmeyer, 1984). The number-average 12

molecular weight (NAMW) can be obtained from the moment equations by noting that Mn is simply: Total mass of polymer = 1 MWmono V Mn = Total (10) moles of polymer V where MWmono is the molecular weight of the repeating monomer unit, which in our case is styrene (?CH2CH (C6H5 )?) with MWstyrene = 104:15g/mol. Equations 1 { 5 can be written in the following dimensionless form:
0

dx3 = (q + q + q )(x ? x ) + i m s 3f 3 p p (x3 )x2x5 ? (x3 ? x4 ) d dx4 = q (x ? x ) + (x ? x )] 1 c 4f 4 2 3 4 d x5 = 2f


s

dx1 = q x ? (q + q + q )x ? i 1f i m s 1 d dx2 = q x ? (q + q + q )x ? m 2f i m s 2 d

d d (x3)x1 p p (x3)x2 x5

(11) (12) (13) (14) (15)

where x1 , x2 , x3 , x4 , x5 , and qc are the dimensionless initiator concentration, monomer concentration, reactor temperature, cooling jacket temperature, growing polymer concentration, and cooling jacket owrate, respectively. We demonstrated in Russo and Bequette (1996b) that this type of cooling jacket energy balance holds for recirculating cooling jackets and that the use of multiple cooling jacket compartments does not change the qualitative steady-state input-output behavior. As noted by Aris (1993), if one is interested in examining the e ect of varying reaction kinetics, then it should not be used to scale the time variable. We are interested in varying the residence time (by varying the reactor owrate) and the reaction kinetic parameters, hence time is scaled using a V . The dimensionless variables and parameters are de ned in the nominal reactor residence time, Q t0 Table 3, with values of the corresponding dimensionless parameters shown in Table 4 (for Table 2 conditions). The dimensionless moment equations are: dx6 = 1 (x )x2 ? x (16) d 2 t t 3 5 6 The number-average molecular weight (NAMW) Mn is:

d d (x3 ) x 1 t t (x3)

dx7 = d

p p (x3)x2x5 ? x7

(17) (18)

x7 is the dimensionless number-average molecular weight (NAMW). The ratio of x 6

7 Mn = MWmono x x

13

6 Steady-state Multiplicity Analysis


6.1 Introduction
It was demonstrated in Russo and Bequette (1995, 1996a, 1996b) that incorporating cooling jacket dynamics in the CSTR modeling equations has an important e ect on the open-loop steady-state and dynamic behavior. An important point is that because the polymer chain moment equations given in the previous section are decoupled from the CSTR material and energy balance equations, they do not change the steady-state multiplicity or dynamic behavior of the underlying four-state polymerization model. However, the polymer chain moment equations may exhibit interesting bifurcation behavior in their own right. An analysis of the steady-state multiplicity behavior of the four-state polymerization CSTR model is presented in the following sections. The e ect of a number of di erent parameter variations (such as the dimensionless heat of reaction, dimensionless heat transfer coe cient, dimensionless cooling jacket feed temperature, and the dimensionless cooling jacket owrate) on the multiplicity behavior is considered. These results are similar in nature to those obtained previously with the lower order CSTR model. The structure of the modeling equations for the four-state polymerization CSTR model makes an analytical treatment of the multiplicity behavior substantially harder than the corresponding threestate A ! B CSTR model. Razon and Schmitz (1987) note the analytical di culties associated with studying the behavior of more complicated reaction networks. The steady-state modeling equations could be collapsed to one nonlinear algebraic equation in the three-state CSTR case studied in Russo and Bequette (1995). This allows one to use results from elementary catastrophe theory in order to analyze the multiplicity behavior. In general, the modeling equations can not be \collapsed" to one single nonlinear equation, hence more sophisticated techniques are necessary. Balakotaiah et al. (1985) use a Liapunov-Schmidt reduction procedure to characterize the behavior of more complicated reaction networks that can not be reduced to a single nonlinear algebraic equation. Fortunately, the steady-state modeling equations for the four-state polymerization system can be \collapsed" into a single nonlinear algebraic equation. In particular, note that at steady-state equation 11 can be solved for x1 as a function of x3 . Then, one can substitute x1 into equation 15 to obtain x5 as a function of x3. Next, substitute x5 into equation 12 which gives x2 as a function 14

of x3 . Finally, equation 14 can be solved for x4 as a function of x3. Now, substituting this into equation 13 (at steady-state) gives: c + p p (x3 ) (x3) + (x4f ? x3 ) q h(x3; qc; p) = qm x2f q + q + q + ( x ) ( x ) + i m s p p 3 3 2 qc (qi + qm + qs )(x3f ? x3 ) = 0 (19) where (x3 ) is 2f d d (x3)qi x1f = x5 (20) t t (x3)(qi + qm + qs + d d (x3)) where x3 (dimensionless reactor temperature) is the controlled output, qc (dimensionless cooling jacket owrate) is the manipulated input, and p is a vector of parameters. (x3) = It is interesting to note that this equation is very similar in nature to the equivalent equation for the three-state CSTR model. ) + q (x ? x ) + (x ? x ) qc = 0 h(x2; qc; p) = qx1f q + (x(2x (21) 2f 2 3f 2 2) 2 + qc The essential di erences in these two equations are in the numerator and denominator of the term. In the three-state CSTR case there is a single nonlinearity ( (x2)) while in the fourstate polymerization CSTR case the nonlinearities ( p (x3) (x3)) multiply each other. This makes a complete analytical treatment for the bifurcation analysis much more di cult. Therefore, a combination analytical/numerical bifurcation analysis is performed.
s

6.2 Determination of Global Multiplicity Diagrams


The necessary conditions for output multiplicity, input multiplicity, and the formation of isolas, respectively, are obtained from the implicit function theorem.

Now, we want to determine if it is possible for the polymerization CSTR model to have input @h . The result is: multiplicity or isolas. Consider the value of @q c @h 2 (x4f ? x3) (25) @qc = 2 ( 2 + qc)2 6= 0 15

@h = 0 h = @q c @h = @h = 0 h = @x 3 @qc

@h = 0 h = @x
3

(22) (23) (24)

Therefore, when qc is the manipulated variable then input multiplicities, isolas, transcritical bifurcations, and pitchfork bifurcations cannot occur between x3 and qc . Upon examination of the derivatives of h(x3; qc ; p) with respect to the parameters, it can be determined that qi , qm , and qs are the only parameters that when varied can lead to input multiplicities or isola behavior; these are therefore important disturbance variables. The derivatives of h(x3; qc ; p) with respect to the other parameters remain the same sign. Elementary catastrophe theory allows one to characterize the number of multiple steady-state solutions to equations 11 { 15. The appearance or disappearance of output multiplicities occurs when higher order derivatives of h with respect to x3 are zero.

The highest order singularity (value of k) satisfying equation 26 determines the number of steadystate solutions. A multiplicity of k + 1 solutions exists around the codimension k singular point satisfying equation 26 (Stewart, 1981). As noted previously, the more complicated structure of the polymerization CSTR modeling equations make an analytical treatment di cult. The highest order singularity found numerically corresponds to k = 2, hence up to three di erent steady-state dimensionless reactor temperatures may exist for a given dimensionless coolant owrate. The determination of regions of feasible and infeasible reactor operation are important for feedback control purposes, since one can see directly if the current operating conditions are \feasible" in a steady-state sense. Therefore, the global parameter space is divided into regions with di erent types of multiplicity behavior. However, there are a large number of parameters associated with the polymerization system so it is not clear how to best present the bifurcation results. This current study portrays the singularity loci on the d - 2-D cross section of the parameter space. This follows the development in Russo and Bequette (1995, 1996a, 1996b) in that kinetic and thermodynamic properties of the system are used to classify the behavior. d is the Damkohler number associated with the dissociation of the initiator into the radical species. If one examines equation 19 we see that as d ! 0 no reaction occurs, since there are no radicals present to initiate the the reaction. (dimensionless heat of reaction) is a direct measure of the exothermicity of the reaction. 16

@h = @ 2h = : : : = @ k h = 0 h = @x @xk 3 @x2 3 3 k +1 @ h 6= 0 +1 @xk 3

(26)

The singularity described by equation 26 is strictly not a \point" but a locus of points in the global parameter space. It is possible to construct singularity loci for a number of low dimensional cross sections of the parameter space. However, since we are interested in how changes in process parameters in uence the location and existence of these singularities on the steady-state inputoutput (x3 - qc ) curves, the singularity locus is constructed over the bounds on the dimensionless coolant owrate (qc ).

@h = @ 2h = 0 h = @x 3 @x2 3 qc 2 0; 1)

(27)

We denote d , , and x3 that satisfy the hysteresis condition (equation 27) as d?hyst , hyst , and x3?hyst . The singularity locus described by equation 27 bounds the region of output multiplicity behavior in the parameter space. A codimension 1 fold bifurcates from the bounds on the cooling jacket owrate. This lower order singularity separates the global multiplicity behavior in the d parameter space.

@h = 0 h = @x 3 qc = 0 @h = 0 h = @x 3 qc ! 1

(28)

(29)

Equations 28 and 29 satisfy the \0-disjoint" and \1-disjoint" loci, respectively. In general, equations 28 and 29 can be satis ed for \low" and \high" temperature limit points, as qc approaches zero or in nity, respectively. Hence, there are a total of four curves in the d - space that bifurcate from the singularity locus given in equation 27.

6.3 Global Multiplicity Behavior


Let us consider a simple parametric study in order to understand how changes in d and in uence the steady-state input-output curves. Figure 2 shows the e ect of varying d for a \smaller" value of ; as d is decreased the characteristics of the steady-state input-output (x3 - qc ) curve changes from monotonic to multiple steady-states to a region of low temperature infeasibility. This infeasibility is 17

also seen if one neglects the cooling jacket energy balance in the modeling equations, which results in a three-state model. Figure 3 corresponds to the same conditions as the d = 0:006 line on Figure 2. Notice that over a range of dimensionless cooling jacket temperatures the reactor temperature is predicted to be less than the cooling jacket temperature, which results in the cooling jacket owrate being negative (cf. Figure 2). A negative cooling jacket owrate is physically impossible. The reactor temperature can not be less than the cooling jacket temperature, since the reactor and cooling jacket have equal feed temperatures and the reaction is exothermic. Figure 4 shows the e ect of increasing d for a \larger" value of ; as d is increased the steady-state input-output curve behavior changes from multiple steady-states to a region of high temperature infeasibility. The high temperature limit point (LP) moves toward in nite cooling jacket owrate. Once again, this infeasibility can be seen in a lower dimensional model that neglects the cooling jacket energy balance. Figure 5 corresponds to the same conditions as the d = 0:008 line on Figure 4. Notice that over a range of dimensionless cooling jacket temperatures the cooling jacket temperature is predicted to be less than the cooling jacket feed temperature. This is impossible, since the reaction is exothermic and the feed temperatures of the reactor and cooling jacket are equal. The codimension 2 (cusp) singularity locus (which speci es the set of hysteresis points) and codimension 1 (fold) \0-disjoint" and \1-disjoint" loci (which specify the set of limit points) divide the d - parameter space into ve di erent regions as shown in Figure 6. These regions correspond to ve di erent steady-state input-output curves, shown in Figure 7. The d and values in regions I - V of Figure 7 are given in Table 5. Region I in the d parameter space corresponds to a monotonic relationship between x3 and qc , while in region II output multiplicity exists between x3 and qc (inverse S shaped input-output curve). Regions III and IV exhibit low temperature and high temperature infeasibilities, respectively, corresponding to \0-disjoint" and \1-disjoint" bifurcations. The presence of \0-disjoint" behavior is associated with overcooling of the reactor by the cooling jacket, whereas \1-disjoint behavior occurs when there is not su cient cooling by the cooling jacket (the heat transfer characteristics of the reactor are poor, as seen in large scale industrial reactors). Region V is an overlap of regions III and IV, where low and high temperature infeasibilities exist when the \0-disjoint" and \1-disjoint" loci intersect. The \dimensional" steady-state input-output curve indicated by the + mark in Figure 6 is shown in Figure 8; notice the reactor operation falls in a \high" temperature operating region of region I. Figure 9 is a plot 18

of the steady-state number-average molecular weight (NAMW) Mn as a function of the coolant owrate. Notice that the \high" temperature operating region has resulted in the production of polymer with \low" number-average molecular weight (NAMW). The d - diagram for the nominal dimensionless operating parameters (given in Table 4) is shown in Figure 10; notice the reactor operation falls in region IV. The steady-state input-output (x3 - qc ) curve is shown in Figure 11; this reactor design exhibits a \high" temperature infeasibility region. This input-output curve is shown in \dimensional" form in Figure 12, where the reactor temperature is shown as a function of the cooling jacket owrate. Figure 13 is a plot of the steady-state numberaverage molecular weight (NAMW) Mn as a function of the coolant owrate for the low and high temperature branches. Notice that the \high" temperature infeasibility region shown in Figure 12 has resulted in a corresponding infeasibility region of number-average molecular weights. Values of the NAMW from 8400 to 12000 do not exist under these operating conditions. The numberaverage molecular weight for the upper branch of Figure 13 is shown as a function of the steady-state temperature in Figure 14. The NAMW changes by almost an order of magnitude over a 40 degree Celsius range. The number-average molecular weight for the lower branch of Figure 13 is shown as a function of the steady-state temperature in Figure 15. The NAMW changes by approximately 15% over a 100 degree Celsius range, with a minimum around 115 oC . The number-average molecular weight is plotted versus the corresponding steady-state coolant owrate in Figure 16. Figure 16 shows that there could be potential problems controlling the number-average molecular weight by manipulating coolant owrate solely, due to the input multiplicity behavior in the \low" molecular weight region. Hidalgo and Brosilow (1990) claimed that this reactor could not be operated safely using only the coolant owrate as the manipulated variable for feedback control, but did not present an explanation. It is clear that the problem is that near the operating range of interest (corresponding to x3 0:75 or T 80 oC ) the gain between the dimensionless reactor temperature and the cooling jacket owrate is approaching zero as one moves further into the infeasibility region (cf. Figure 12. Hidalgo and Brosilow (1990) also placed a maximum constraint on the cooling jacket owrate of Qc = 60 liters min ; this corresponds to a maximum dimensionless cooling jacket owrate of qc = 3:8095. Notice from Figures 11 and 12 that this constraint virtually eliminates any ability to operate in the reactor range of interest with only the cooling jacket owrate. 19

7 In uence of Polymerization CSTR Parameters on the Multiplicity Behavior


7.1 Introduction
It is clear from the results presented in the previous section that the physical constraints imposed on the cooling jacket owrate lead to interesting multiplicity behavior. We saw that incorporating more complicated kinetics did not change the nature and types of multiplicity diagrams that occur. The multiplicity behavior shown in Figures 2 and 3 is now easily explained. As d is decreased in Figure 2, we move from region I through region II into region III. As d is increased in Figure 3, we move from region II into region IV. The focus of this section is to explore the in uence of process design or operation changes on the bifurcation behavior of the four-state polymerization CSTR model. The number of process parameters used to describe the dimensionless polymerization model makes a complete parametric study impractical; however, by using the multiplicity results of the previous section we are able to demonstrate the more relevant parametric e ects.

7.2 E ect of Reactor Scaleup


Let us consider the e ect of reactor sizing on the hysteresis locus behavior. In Russo and Bequette (1995) we showed that when the cooling jacket feed temperature was equal to the reactor feed temperature, the Damkohler number along the hysteresis locus ( hyst ) is a constant. The dimensionless heat of reaction along the hysteresis locus ( hyst ) changes by a factor of q+ q (recall that is the dimensionless heat transfer coe cient and q is the dimensionless owrate).
hyst (qc ! 1) =

q+ q

hyst (qc = 0)

(30)

Hence, in this case we can clearly see how the multiplicity behavior changes as a function of the dimensionless heat transfer coe cient ( ). If the residence time is held constant upon reactor 1 scaleup, then Dr , where Dr is the diameter of the reactor. Therefore, a \larger" reactor will have a lower value of , resulting in a shorter hysteresis locus. This larger reactor has less e cient cooling characteristics, which can shift the reactor operation into an undesirable region. 20

We compare the d - diagrams for two di erent reactor systems in order to demonstrate that the polymerization CSTR system exhibits the same type of dependence on dimensionless heat transfer coe cient (when the cooling jacket feed temperature is equal to the reactor feed temperature). Figure 6 corresponds to a 3000 liter reactor volume ( = 0:74074) whereas Figure 17 has a reactor volume of 1500 liters ( = 0:93328). As one can see, the larger reactor has a shorter hysteresis locus. It is clear that if we scale down the reactor size \enough" the reactor operation will move from region I through region IV into region II, resulting in multiple steady-state behavior. The multiplicity behavior in this case is a result of the \low" temperature operating region becoming feasible as the heat transfer characteristics improve when we \scaledown" the reactor size. An important point is that the hysteresis locus length is once again related to the value of the dimensionless heat transfer coe cient ( ) and the process owrates. Table 6 shows the values of hyst (qc = 0) and hyst (qc ! 1) for a number of process owrates and dimensionless heat transfer coe cients (the unspeci ed parameters are the default values given in Table 4). If we examine hyst(qc !1) we see that hyst(qc=0) hyst (qc ! 1) = qi + qm + qs + (31) qi + qm + qs hyst (qc = 0) which is the same relationship that was seen in the simpler three-state CSTR model presented in Russo and Bequette (1995).

7.3 Positive Exponential Approximation


A common assumption in steady-state multiplicity studies is the \so-called" positive exponential approximation. This approximation has been used by a number of researchers (Farr and Aris, 1986). The beauty in using the positive exponential approximation is that it often enables one to obtain an analytical expression for characterizing the multiplicity behavior. However, it was shown in Russo and Bequette (1995) that this approximation tends to overestimate the multiplicity area in the global parameter space. Generally, the positive exponential approximation consists of allowing the dimensionless activation energy terms to become \large" (in the limit sense) with respect to the reactor temperature. This simpli es the dimensionless Arrhenius temperature dependence of the reaction rates, given by E , Et p (x3), t (x3 ), and d (x3) (shown in Table 3). However, recall that d = Ed p t = Ep , and 21

hence it does not make physical sense to allow these dimensionless activation terms to become large. Consider instead a situation where the individual activation energies become large with respect to the reactor temperature. In this case, we see that d ! 1, t ! 1, and p ! 1. The resulting d - diagram is shown in Figure 18. The multiplicity behavior is grossly overestimated and shifted in the parameter space, as compared to Figure 6. The reactor operation is predicted to lie in region III whereas the operation exists in region I. We see that this form of the positive exponential approximation obscures the true nature of the multiplicity behavior.

Ep p = RTf 0 ,

7.4 E ect of Process Flowrates on the Multiplicity Behavior


In addition to studying the e ect of reactor size it is also important to consider how process operation a ects the multiplicity behavior. In particular, it is interesting to understand how the process owrate in uences the bifurcation behavior. In Russo and Bequette (1995) we saw that when the cooling jacket feed temperature is equal to the reactor feed temperature, then the value of the Damkohler number along the hysteresis locus ( hyst ) was linearly proportional to the process owrate (q ). Therefore, if the reactor operation is in a region of multiple steady-states, one could simply decrease the process owrate such that hyst < , which will move the reactor operation region into region I, where the steady-state input/output curve is monotonic. In the case when the cooling jacket and reactor feed temperatures are not equal, then one has to decrease the process owrate such that max qc ( hyst) < in order to assure a monotonic steady-state hyst input-output curve. However, for the three-state CSTR it is easy to show @ @ hyst > 0, as long as the reactor feed temperature was greater than the cooling jacket feed temperature, hence the maximum hyst occurs as qc ! 1. When the reactor feed temperature is less than the cooling jacket feed hyst temperature, @ @ hyst < 0, so the maximum hyst occurs as qc ! 0. We are interested in seeing if these same relationships hold for the four-state polymerization CSTR model. Notice that these conditions on the Damkohler number are su cient conditions; there exists region I behavior (monotonic steady-state input-output curves) in the parameter space when d?hyst > d (cf. Figures 6, 10, 17, etc.). However, if d?hyst < d then we are guaranteed monotonic steadystate input-output curve behavior (no output multiplicities). For the purposes of this case study, we choose the cooling jacket feed temperature equal to the 22

reactor feed temperature. We want to understand how the various process owrates (initiator, monomer, and solvent) in uence the multiplicity behavior. The Damkohler number along the hysteresis locus ( d?hyst ) is not simply a linear function of the owrate as it was in the threestate CSTR model. For example, notice the behavior of d?hyst in Figure 19. We see that as one varies the initiator owrate (qi ) that d?hyst initially decreases and then it levels out and starts increasing (for a xed value of the solvent owrate qs ). This behavior is unique to the four-state polymerization CSTR model, in that there are multiple values of the initiator owrate for the same hysteresis Damkohler number. There exist certain operating regimes in which multiple steady-states can be avoided by either increasing the initiator owrate \enough" (thus decreasing the residence time of the reactor for a lower conversion), or by decreasing the initiator owrate \enough" (thus e ectively \quenching" the reaction due to an absence of radicals). In addition, notice on Figure 19 that decreasing the monomer owrate (qm ) tends to decrease d?hyst . Therefore, one can remove multiple steady-state behavior by decreasing the monomer owrate. This gure allows one to nd process operating conditions to guarantee a monotonic steady-state input-output curve over a range of operating conditions. For example, consider a Damkohler number of d = 0:015. For monomer owrates of qm = 0:3 and below (with a xed solvent owrate of qs = 0:48571) we see that there exists a wide range of initiator owrates which guarantee operation with a monotonic steady-state input-output curve. It is well-known that the amount of initiator present in the system generally controls the polymer properties (Hidalgo and Brosilow, 1990), hence the initiator owrate may only be varied over a narrow range to obtain the desired polymer properties. We see in Figure 20 that d?hyst initially decreases then starts increasing as one increases the initiator owrate (qi ) for a xed value of qm . We can set the desired initiator owrate and nd from Figure 20 the solvent owrate which guarantees monotonic steady-state input-output behavior. For example, for a value of d = 0:0175 and qi = 0:1, values of the solvent owrate qs = 0:4 and lower satisfy the monotonicity conditions. Figures 21 and 22 represent cases where the initiator owrate is xed and the monomer (qm ) and solvent (qs ) owrates were varied. We see in these cases that the hysteresis Damkohler number is a monotonic increasing function of both the solvent and the monomer owrates. Consider a polymerization CSTR with d = 0:0175 and a monomer owrate of qm = 0:4. We see in Figure 21 that we are guaranteed monotonic steady-state input-output behavior only if the solvent owrate is 23

lower than qs 0:5. However, if the monomer owrate is decreased from qm = 0:4 to qm = 0:2 then monotonic steady-state input-output behavior is possible with larger values of the solvent owrate (qs 0:7 and lower).

7.5 E ect of Process Design and Operation Changes


We want to explore the e ect of process design or operation changes on the bifurcation behavior of the polymerization CSTR model. The system parameters given in Hidalgo and Brosilow (1990) result in the dimensionless parameters shown in Table 4. The d - diagram for this parameter set was shown in Figure 10. The reactor operation fell in region IV (\high" temperature infeasibility region); the corresponding input-output (x3 - qc ) curve was shown in Figure 11. Let us rst consider a change in the reactor size. If we decrease the reactor volume from 3000 liters ( = 0:74074) to 1500 liters ( = 0:93328), the reactor operation shifts from region IV (Figure 10) into region II (associated with steady-state multiplicity), as shown in Figure 23. Although this design obviously su ers from multiple steady-state behavior (shown in Figure 24), we now are able to operate over the entire range of desired reactor temperatures. However, the presence of output multiplicities in the open-loop steady-state output curve means that we will need to use feedback control to guarantee stable operation. Now, consider a change in the cooling jacket feed temperature for a 3000 liter reactor. If we decrease the cooling jacket feed temperature from x4f = ?1:14320 to x4f = ?1:5, the reactor operation shifts from region IV to region II, as shown in Figure 25. Once again, we have multiple steady-states, shown in Figure 26, which necessitates the use of feedback control to stabilize the system. However, we have removed the \high" temperature infeasibility region. Finally, we re-examine Figure 6. We see that by increasing the cooling jacket feed temperature (from x4f = ?1:14320 to x4f = 0:0) that the reactor operation has shifted from region IV to region I (monotonic steady-state input-output curve). However, we lose the ability to operate the reactor in a \low" temperature operating region (cf. Figure 8).

24

8 Conclusions
We have shown in this work that the bifurcation analysis shown in Russo and Bequette (1995, 1996a, 1996b) could be applied to a jacketed styrene polymerization CSTR. The role of physical constraints on the steady-state input-output behavior of a styrene polymerization model was studied. In particular, an analysis of the steady-state multiplicity behavior was conducted, and the e ect of dimensionless values of the heat of reaction, overall heat transfer coe cient, cooling jacket feed temperature, and physical limits on the cooling jacket owrate were considered. It was shown in Russo and Bequette (1996b) that this type of cooling jacket energy balance holds true for recirculating cooling jackets and that the use of multiple cooling jacket compartments does not change the qualitative steady-state input-output behavior. The global multiplicity behavior was characterized in terms of a Damkohler number for initiator decomposition and the dimensionless heat of reaction. The presence of low and high temperature infeasibility regions associated with disjoint bifurcation behavior was demonstrated. Polymer chain moment equations were appended to the four-state polymerization model. These equations were used to calculate the number-average molecular weight (NAMW). The addition of these equations does not change the underlying steady-state multiplicity or dynamic behavior of the polymerization CSTR. However, these polymer chain moment equations may exhibit interesting behavior of their own. For example, when the polymerization CSTR has a high temperature infeasibility region, there exists a range of infeasible number-average molecular weights. A polymerization CSTR with solely a high temperature operating region (e.g. the cooling jacket feed temperature may be too high) results in the production of polymer with a low number-average molecular weight (NAMW). Polymers with a low NAMW may exhibit input multiplicity with respect to cooling jacket owrate and thus may be di cult to control. The e ect of process design and operation parameters on the bifurcation behavior was considered. In particular, when the cooling jacket feed temperature is equal to the reactor feed temperature, the hysteresis locus length (which speci es the di erent types of input-output behavior) was a function of the process owrates and the dimensionless heat transfer coe cient. The positive exponential approximation was shown to obscure the true nature of the multiplicity behavior. The hysteresis Damkohler number ( d?hyst ) was not a linear function of the process owrates, unlike 25

the A ! B reaction network case. However, the in uence of the various process owrates on the Damkohler number hysteresis locus was determined in order to guarantee monotonic steady-state input-output behavior. We saw that under certain conditions a 3000 liter CSTR exhibited a region of high temperature infeasibility which was not present when the reactor was scaled down to 1500 liters. Decreasing the cooling jacket feed temperature was also found to move the steady-state input-output behavior into a globally feasible operating region.

Acknowledgment
This work was partially supported by a grant from the Petroleum Research Foundation, administered by the American Chemical Society. A portion of this work was performed while Wayne Bequette was on sabbatical at Northwestern University.

26

References
Adebekun, A.K., K.M. Kwalik, and F.J. Schork (1989). Steady-state Multiplicity During Solution Polymerization of Methyl Methacrylate in a CSTR. Chem. Eng. Sci., 44, 2269-2281. Aris, R. (1993). Ends and Beginnings in the Mathematical Modelling of Chemical Engineering Systems. Chem. Eng. Sci., 48, 2507-2517. Balakotaiah, V., D. Luss, and B.L. Key tz (1985). Steady-state Multiplicity Analysis of Lumpedparameter Systems Described by a Set of Algebraic Equations. Chem. Eng. Comm., 36, 121-147. Billmeyer, F.W. (1984). Textbook of Polymer Science., Wiley and Sons. Brooks, B.W. (1981). Dynamic Behaviour of a Continuous- ow Polymerisation Reactor. Chem. Eng. Sci., 36, 589-593. Choi, K.Y. (1986). Analysis of Steady State of Free Radical Solution Polymerization in a Continuous Stirred Tank Reactor. Polym. Eng. Sci., 26, 975-981. Congalidis, J.P., J.R. Richards, and W.H. Ray (1989). Feedforward and Feedback Control of a Solution Copolymerization Reactor. AIChE J., 35, 891-907. Doedel, E.J. (1986). AUTO: Software for Continuation and Bifurcation Problems in Ordinary Di erential Equations., California Institute of Technology, Pasadena, CA. Farr, W.W. and R. Aris (1986). \Yet who would have thought the old man to have had so much blood in him?" - Re ections on the Multiplicity of Steady States of the Stirred Tank Reactor. Chem. Eng. Sci., 41, 1385-1402. Hamer, J.W., T.A. Akramov, and W.H. Ray (1981). The Dynamic Behavior of Continuous Polymerization Reactors - II. Nonisothermal Solution Homopolymerization and Copolymerization in a CSTR. Chem. Eng. Sci, 36, 1897-1914. Hidalgo, P.M. and C.B. Brosilow (1990). Nonlinear Model Predictive Control of Styrene Polymerization at Unstable Operating Points. Comp. Chem. Eng., 14, 481-494.

27

Jaisinghani, R. and W.H. Ray (1977). On the Dynamic Behavior of a Class of Homogeneous Continuous Stirred Tank Polymerization Reactors. Chem. Eng. Sci., 32, 811-825. Kim, K.J., K.Y. Choi, and J.C. Alexander (1991). Dynamics of a Cascade of Two Continuous Stirred Tank Polymerization Reactors with a Binary Initiator Mixture. Polym. Eng. Sci., 31, 333-352. Pinto, J.C. and W.H. Ray (1995a). The Dynamic Behavior of Continuous Solution Polymerization Reactors - VII. Experimental Study of a Copolymerization Reactor. Chem. Eng. Sci., 50, 715-736. Pinto, J.C. and W.H. Ray (1995b). The Dynamic Behavior of Continuous Solution Polymerization Reactors - VIII. A Full Bifurcation Analysis of a Lab-scale Copolymerization Reactor. Chem. Eng. Sci., 50, 1041-1056. Ray, W.H. (1993). Modeling and Control of Polymerization Reactors. In Dynamics and Control of Chemical Reactors, Distillation Columns, and Batch Processes (DYCORD+ 1992), J.G. Balchen (Ed.), Pergamon Press, 161-170. Razon, L.F. and R.A. Schmitz (1987). Multiplicities and Instabilities in Chemically Reacting Systems - a Review. Chem. Eng. Sci., 42, 1005-1047. Russo, L.P. and B.W. Bequette (1995). Impact of Process Design on the Multiplicity Behavior of a Jacketed Exothermic CSTR, AIChe J., 41, 135-147. Russo, L.P. and B.W. Bequette (1996a). E ect of Process Design on the Open-Loop Behavior of a Jacketed Exothermic CSTR, Comp. Chem. Eng., 20, 417-426. Russo, L.P. and B.W. Bequette (1996b). Operability of Chemical Reactors: Multiplicity Behavior of Extended CSTR Models, AIChE J., Submitted July 1996. Schmidt, A.D., A.B. Clinch, and W.H. Ray (1984). The Dynamic Behavior of Continuous Polymerization Reactors - III. An Experimental Study of Multiple Steady-states in Solution Polymerization. Chem. Eng. Sci., 39, 419-432. Schmidt, A.D. and W.H. Ray (1981). The Dynamic Behavior of Continuous Polymerization Reactors - I. Isothermal Solution Polymerization in a CSTR. Chem. Eng. Sci., 36, 1401-1410. 28

Stewart, I. (1981). Applications of Catastrophe Theory to the Physical Sciences. Physica, 2D, 245-305. Teymour, F. and W.H. Ray (1989). The Dynamic Behavior of Continuous Polymerization Reactors - IV. Dynamic Stability and Bifurcation Analysis of an Experimental Reactor. Chem. Eng. Sci., 44, 1967-1982. Teymour, F. and W.H. Ray (1992). The Dynamic Behavior of Continuous Polymerization Reactors - V. Experimental investigation of Limit-cycle Behavior for Vinyl Acetate Polymerization. Chem. Eng. Sci., 47, 4121-4132. Teymour, F. and W.H. Ray (1992). The Dynamic Behavior of Continuous Polymerization Reactors - VI. Complex Dynamics in Full-scale Reactors. Chem. Eng. Sci., 47, 4133-4140.

29

Table 1: Variables in the styrene polymerization model. f - Initiator e ciency Ci - Initiator concentration Cm - Monomer concentration Cgp - Growing polymer concentration T - Reactor temperature Tc - Cooling jacket temperature 0 - Concentration of dead polymer 1 - Concentration of monomer units present as polymer Hr - Heat of reaction - Density of the reaction mixture - Density of the cooling jacket mixture c Cp - Heat capacity of the reaction mixture Cpc - Heat capacity of the cooling jacket mixture U - Overall heat transfer coe cient A - Heat transfer area kd - Rate constant for initiator decomposition kp - Rate constant for propagation kt - Rate constant for termination t - time Qi - Initiator owrate Qm - Monomer owrate Qs - Solvent owrate Qt - Total owrate through reactor Qc - Cooling jacket owrate V - Reactor volume Vc - Cooling jacket volume Cif - Initiator feed concentration Cmf - Monomer feed concentration Tf - Reactor feed temperature Tcf - Cooling jacket feed temperature 30

Table 2: Representative values of the styrene polymerization parameters (taken from Hidalgo and Brosilow, 1990).

f kd0 Ed kt0 Et kp0 Ep

= 0.6 = 5.95 * 1013 s?1 = 123,855.658 J mol?1 = 1.25 * 109 s?1 = 7008.702 J mol?1 = 1.06 * 107 l mol?1 s?1 = 29572.898 J mol?1

Qi Qs

= 0.03 l s?1 = 0.1275 l s?1

Qm = 0.105 l s?1 Qc = 0.131 l s?1 V Vc


= 3000 l = 3312.4 l

Cif = 0.5888 mol l?1 Cmf = 8.6981 mol l?1 Tf


= 330 K

? Hr = 69,919.56 J mol?1
UA Cp
c Cpc

= 293.076 J s?1 K?1 = 1507.248 J l?1 K?1 = 4045.7048 J l?1 K?1

Tcf = 295 K

31

Table 3: Dimensionless variables and parameters for the four-state polymerization CSTR model (Qt0 = Qio + Qm0 + Qs0 ).

x1 x3 x6 qi qs x1f x3f
d p
1

i = CC mf 0

x2 x4 x7 qm qc x2f x4f
t

m = CC mf 0

Tf 0 = T? Tf 0 p
0 = Cmf 0

?Tf 0 = TcT f0 p
1 = Cmf 0

Qi = Q t0 Qs = Q t0
if = CC mf 0

Qm = Q t0 Qc = Q t0 Cmf = C mf 0 Tf 0 = TcfT? p f0 Et = E p

?Tf 0 = TfT f0 p
d = E Ep

Ep = RT f0 V = V c
2

= =

(? Hr )Cmf 0 p Cp Tf 0

Cp c Cpc

=
t

UA Cp Qt0

d p d (x3 ) p (x3)

V kd0 exp(? d p ) = Q t0
mf 0 = VC Qt0 kp0exp(? p ) dx = exp( 1+ x3 3 ) p

mf 0 = VC Qt0 kt0exp(? t p) t0 = Q V t

t (x3)

tx = exp( 1+ x3 3 ) p gp = CC mf 0

x3 = exp( 1+ x3 )
p

x5

32

Table 4: Dimensionless four-state polymerization model parameters values.

qi

= 0.11429

qm = 0.4 x1f = 0.06769 x3f = 0.0


d p
1

qs = 0.48571 x2f = 1.0 x4f = -1.14320


t

= 4.18808 = 10.77879 = 0.90569 = 0.74074

= 0.23699 = 13.17936

= 0.37256 = 0.01688 = 2.19560e7


t

d p

= 9.6583e12

Table 5: d and values used in Figure 7 (x4f = 0:0). region d I 0.018 8 II 0.01 8 III 0.003 8 IV 0.004 14 V 0.001 14 33

Table 6: Representative values of hyst (qc = 0) and hyst (qc ! 1) for a number of di erent process parameters. hyst(qc!1) Parameter hyst (qc = 0) hyst (qc ! 1) hyst(qc =0) = 0.74074 4.86712 8.47244 1.74074 = 0.93328 4.86712 9.40955 1.93328 qi = 0.15 4.88992 8.38717 1.71520 qm = 0.3 5.65829 10.31529 1.82304 qs = 0.4 4.33069 7.83932 1.81018

34

Appendix 1
Derivation of the Polymerization Kinetic Expressions
Reaction rates of the free-radical polymerization kinetics (Hidalgo and Brosilow, 1990):

rI rR rP rM

= ?kd Ci = 2fkd Ci ? ki Cm Cr
2 = ki Cm Cr ? ktCgp

(32) (33) (34) (35)

= ?ki CmCr ? kp Cm Cgp

The long chain hypothesis is made to the monomer reaction rate, i.e. ki Cm Cr kpCm Cgp, because most of the polymer consumed is added to existing polymer chains and not in forming new polymer chains. In this case, rM becomes: rM = ?kpCmCgp (36) The quasi-steady-state approximation is made to the radical species R, P , since the radical species lifetimes are short compared to the system time constants (Jaisinghani and Ray, 1977). It is also assumed that no radicals exit the reactor. In this case, the radical reaction rates are set to zero (quasi-steady-state approximation), i.e. rR 0, rP 0. Solving these equations allows us to obtain the expression for the growing polymer chain concentration Cgp . 2fkd Ci = ki Cm Cr
2 kiCmCr = ktCgp

(37) (38)

By equating these two equations we obtain:


2 = 2fkdCi ktCgp

(39)

Solving for the growing polymer concentration, Cgp:

Cgp =

2fkd Ci

kt

(40)

35

Appendix 2
Derivation of the Polymer Chain Moment Equations
From Schmidt and Ray (1981), the two polymer moment equations are written as:

Here, 0 is the total molar concentration of dead polymer and 1 is the total molar concentration of monomer units present as polymer (Schmidt and Ray, 1981) and is:

d 0 = 1 k C 2 ? Qt dt 2 tc gp V 0 d 1 = 1 k C 2 ? Qt dt 1 ? tc gp V

(41)
1

(42)

Cm = k C kp+ p m ktc Cgp


Substituting into the equation for ddt1 we obtain:

(43)

d 1 = k C 2 + k C C ? Qt (44) tc gp p m gp V 1 dt 2 . From the long chain However, from the quasi-steady-state approximation: ki Cm Cr = ktc Cgp hypothesis we note that: ki Cm Cr kpCm Cgp. Therefore, under these assumptions
2 ktcCgp

kpCmCgp

(45)

So, the second polymer moment equation becomes:

d 1 = k C C ? Qt p m gp V dt

(46)

36

monomer flowrate monomer feed conc feed temp

Qm Cmf Tf

feed temp

solvent Tf Q s flowrate

Q i initiator flowrate initiator C if feed conc Tf feed temp Tcf Qc


coolant feed temp coolant flowrate

coolant temp

Tc

C i (initiator conc) C m (monomer conc) T (reactor temp) Tc (coolant temp)

Qt total flowrate C i initiator conc C m monomer conc T reactor temp

Figure 1: Process ow diagram for the jacketed styrene polymerization CSTR.

1.7 1.5 1.3 1.1 d = 0.018 d = 0.015 d = 0.012 d = 0.009 d = 0.006

x3

0.9 0.7 0.5 0.3 0.1 0.0 0.1 0.2

0.3

qc

Figure 2: Steady-state dimensionless reactor temperature versus cooling jacket owrate, x4f = 0:0. 37

= 6,

1.6 1.4 1.2 1.0

x3

0.8 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6

x3 > x4 x3 < x4

x4

Figure 3: Steady-state dimensionless reactor temperature versus cooling jacket temperature, d = 0:006, = 6, x4f = 0:0.
3.5 3.0 2.5 2.0 d = 0.008 d = 0.0065 d = 0.00625 d = 0.006

x3
1.5 1.0 0.5 0.0 0 5 10 15 20

qc

Figure 4: Steady-state dimensionless reactor temperature versus cooling jacket owrate, = 10, x4f = 0:0. 38

2.0

1.5

x3

1.0

0.5

x4 > x4f x4 < x4f

0.0 0.0 0.2 0.4 0.6 0.8

x4

Figure 5: Steady-state dimensionless reactor temperature versus cooling jacket temperature, d = 0:008, = 10, x4f = 0:0.

20.0x10-3

I
16.0 12.0 Hysteresis locus "0-disjoint" loci "-disjoint" loci

II
8.0 4.0

III
0.0 4 6 8 10

IV
12 14 16 18 20

Figure 6: Global multiplicity behavior for the styrene polymerization CSTR model, x4f = 0:0, = 0:74074. 39

2.6 2.2 1.8 1.4 1.0 0.6 0.2 0 1 2 3 4 2.8 2.4 2.0 1.6
x3

2.6

2.2 1.8
x3

II

x3

1.4 1.0 0.6 0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

qc

qc

III

1.2 0.8 0.4 0.0 0.0 0.1 0.2 0.3 0.4 5 0.5 0.6

qc
4

IV
4

3
3
x3 x3

1 0

0 0 5 10 15 20 25 30

qc

10

15

20

25

qc

Figure 7: Bifurcation diagrams for the styrene polymerization CSTR.

40

210

190

T (oC)

170

150

130 0 20

Qc (liters/min)

40

60

80

Figure 8: Steady-state reactor temperature versus cooling jacket owrate, d = 0:01688, 13:17936, x4f = 0:0.
8400 8200

NAMW (g/mol)

8000 7800 7600 7400 7200 0 20

Qc (liters/min)

40

60

80

Figure 9: Number-average molecular weight (NAMW) versus cooling jacket owrate, d = 0:01688, = 13:17936, x4f = 0:0. 41

0.25 0.20 Hysteresis locus "0-disjoint" loci "-disjoint" loci

I
0.15

d
0.10

II
0.05 0.00 4 6 8 10 12

IV
14 16 18 20

Figure 10: Global multiplicity behavior for the styrene polymerization CSTR model, x4f = ?1:14320.
3.0 2.5 2.0 1.5

x3
1.0 0.5 0.0 -0.5 0 2 4 6 8 10 12 14

qc

Figure 11: Steady-state dimensionless reactor temperature versus cooling jacket owrate, d = 0:01688, = 13:17936, x4f = ?1:14320. 42

160 140 120

T (oC)

100 80 60 40 0 50 100 150 200

Qc (liters/min)
Figure 12: Steady-state reactor temperature versus cooling jacket owrate, d = 0:01688, 13:17936, x4f = ?1:14320.
3

80x10

70

NAMW (g/mol)

60 50 40 30 20 10 0 0 50 100 150 200

Qc (liters/min)
Figure 13: Number-average molecular weight (NAMW) versus cooling jacket owrate, d = 0:01688, = 13:17936, x4f = ?1:14320. 43

80x10

70

NAMW (g/mol)

60 50 40 30 20 10 40 50 60 o 70 80

T ( C)

Figure 14: Number-average molecular weight (NAMW) versus steady-state reactor temperature: low temperature operating region, d = 0:01688, = 13:17936, x4f = ?1:14320.
8400 8200

NAMW (g/mol)

8000 7800 7600 7400 7200 7000 100 120 140

T ( C)
Figure 15: Number-average molecular weight (NAMW) versus steady-state reactor temperature: high temperature operating region, d = 0:01688, = 13:17936, x4f = ?1:14320. 44

160

180

200

8400 8200

NAMW (g/mol)

8000 7800 7600 7400 7200 7000 0 50 100 150 200

Qc (liters/min)
Figure 16: Number-average molecular weight (NAMW) versus cooling jacket owrate: high temperature operating region, d = 0:01688, = 13:17936, x4f = ?1:14320.

20.0x10

-3

I
16.0 12.0

Hysteresis locus "0-disjoint" loci "-disjoint" loci

II
8.0 4.0

III
0.0 4 6 8 10 12

IV
14 16 18 20

Figure 17: Global multiplicity behavior for the styrene polymerization CSTR model, x4f = 0:0, = 0:93328. 45

40.0x10-3 35.0 30.0 25.0

Hysteresis locus "0-disjoint" loci "-disjoint" loci

20.0 15.0 10.0 5.0 0.0 10 15

II IV III
20

25

30

Figure 18: Global multiplicity behavior for the styrene polymerization CSTR model using the positive exponential approximation, t ! 1, d ! 1, p ! 1, x4f = 0:0.

50x10-3

40

d-hyst

qm = 0.2 qm = 0.3 qm = 0.4 qm = 0.5 qm = 0.6

30

20

10 0.2 0.4 0.6 0.8 1.0

qi

Figure 19: Damkohler number hysteresis locus as a function of the initiator owrate for various values of the monomer owrate, x4f = 0:0, qs = 0:48571. 46

70x10-3 60 50

d-hyst

qs = 0.2 qs = 0.4 qs = 0.6 qs = 0.8 qs = 1.0

40 30 20 10 0.2 0.4 0.6 0.8 1.0

qi

Figure 20: Damkohler number hysteresis locus as a function of the initiator owrate for various values of the solvent owrate, x4f = 0:0, qm = 0:4.

70x10-3 60 50

d-hyst

40 30 20 10

qs = 0.2 qs = 0.4 qs = 0.6 qs = 0.8 qs = 1.0

0.0

0.2

0.4

0.6

0.8

qm

Figure 21: Damkohler number hysteresis locus as a function of the monomer owrate for various values of the solvent owrate, x4f = 0:0, qi = 0:11429. 47

60x10-3 50 40 qm = 0.2 qm = 0.3 qm = 0.4 qm = 0.5 qm = 0.6

d-hyst

30 20 10 0

0.2

0.4

0.6

0.8

1.0

qs

Figure 22: Damkohler number hysteresis locus as a function of the solvent owrate for various values of the monomer owrate, x4f = 0:0, qi = 0:11429.

0.35 0.30 0.25 Hysteresis locus "0-disjoint" loci "-disjoint" loci

0.20 0.15 0.10 0.05 0.00 4 6 8 10 12

II

IV

14

16

18

20

Figure 23: Global multiplicity behavior for the styrene polymerization CSTR model, V = 1500 liters, = 0:93328, x4f = ?1:14320. 48

x3

0 0.0 0.5 1.0 1.5 2.0 2.5 3.0

qc

Figure 24: Steady-state dimensionless reactor temperature versus cooling jacket owrate, V = 1500 liters, d = 0:01688, = 13:17936, = 0:93328, x4f = ?1:14320.

0.6 0.5 0.4 Hysteresis locus "0-disjoint" loci "-disjoint" loci

I
d
0.3 0.2 0.1 0.0 4 6 8 10 12

II IV

14

16

18

20

Figure 25: Global multiplicity behavior for the styrene polymerization CSTR model, x4f = ?1:5. 49

3.4 2.6 1.8

x3
1.0 0.2 -0.6 0 1 2 3 4 5

qc

Figure 26: Steady-state dimensionless reactor temperature versus cooling jacket owrate, d = 0:01688, = 13:17936, x4f = ?1:5.

50

Potrebbero piacerti anche