Modeling of Polyvinyl Acetate Polymerization Process for Control Purposes

18 th European Symposium on Computer Aided Process Engineering ESCAPE 18 Bertrand Braunschweig and Xavier Joulia (Editors) 2008 Elsevier B.V./Ltd. All rights reserved. Modeling of Polyvinyl Acetate Polymerization Process for Control Purposes Teodora Miteva a, odrigo Alvarez b, Nadja Hvala a, Dolores Kukanja c a Jožef Stefan Institute, Jamova 39, SI-1000 Ljubljana, Slovenia b UPC, Enginyeria Química, Avda. Diagonal 647, Pab. G-2, E-08028 Barcelona, Spain c MITOL, Sežana, Partiyanska cesta 78, SI 6210, Slovenia Abstract The research work presented in this paper considers the polymerization process in MITOL factory in Sežana (Slovenia). The challenge is to increase the production rate, while keeping quality parameters at the desired values. For this purpose a model based on differential and algebraic equations (DAE) has been already constructed 7. The model is intended to take over the role of a soft sensor, being able to provide estimates of the given process outputs. In this paper we derive an extended version of the model in order to reach fully predictive capabilities. By using modeling from first principles (energy balance), the dynamics of the temperature is derived as a function of the concentrations of reacting chemicals. The preliminary simulation results based on real plant data, derived in gpoms environment, are presented. Keywords: vinyl acetate, control, modeling, gpoms 1. Introduction Emulsion polymerization of vinyl acetate is quite a complex chemical process. Efforts have been made to determine the mechanisms of the reaction. Initial estimations of the number of particles and conversion, based upon the Smith and Ewarts polymerization model were written by O Donnel and Mesrobian 1. Dunn and Taylor 2 went further and developed many of the theories commonly accepted nowadays. In the literature it is also possible to find a good number of comprehensive models of the vinyl acetate polymerization reaction, gathering the different theories and mechanisms proposed. In 1992, Urquiola et al. 3,4,5 wrote a series of papers about the polymerization of vinyl acetate using a polymerizable surfactant. The presented work addressis modeling of a polymerization process in a chemical factory Mitol in Sežana, Slovenia. The problem we have is to design optimal control actions required to reach the final product with specified quality parameters in as short time as possible. The motivation arises from the fact that current batch control is done by following a recipe derived from experience. The large quantities of raw materials and equipment resources involved in the batch prevent us from performing experiments directly on the plant. Therefore we need a good predictive model of the polymerization reaction. The model of Aller et al. 6,7 was designed to provide estimates of four outputs: conversion, particles size, solids content and viscosity, having the temperature and initial amounts and flow rates of the monomer and initiator as model inputs. The problem appeared that having the temperature as a model input does not allow us to fully optimize the polymerization process and to observe the changes of the temperature, and consequently the process outputs, by supplementing different amounts

2 T. Miteva of monomer and initiator. The sufficient amount of initiator was also difficult to be predicted. In our work we extend the proposed model with energy balance equations and in this way simulate the temperature in the reactor as an additional model variable. In this way the necessary control actions for shortening the batch duration could be studied. 2. Process description The modeled reaction is polymerization of vinyl acetate using potassium persulfate (KPS) as initiator and polyvinyl alcohol as protective colloid. The production process is semi batch polymerization. Initial amounts of initiator and monomer are added into the system as well as the whole amount of the polyvinyl alcohol. The reactor is closed and the heating starts. When the temperature of the reactor reaches approximately 70 C, the operator stops the hot water through the heating (Figure 1). The exothermic reaction and the heat of the remaining water filling the heating coat continue to rise the temperature over the set point. After the reactor temperature reaches a certain level, the remaining monomer starts to be pumped into the reactor with a continuous flow of 370 kg/h. Figure 1. Scheme of polymerization reactor in MITOL The monomer flow rate is controlled by a PID controller acting on the pump. In the beginning of the reaction, the temperature can reach 82 C. From now on, for the product quality reasons, it is necessary to keep the temperature between 75-80 C. Temperature control is performed by manually adding a small amount of initiator every time the temperature decreases. The temperature of the added monomer is the outhdoors temperature. When all the monomer is added into the system, a larger amount of initiator is added in order to terminate the reaction. The temperature is then allowed to increase up to 90 C. The reaction is considered as finished when the temperature starts to decrease again. The heat in the reactor is released in three ways: Direct cooling in the reactor. Losses to the surroundings Heating the incoming flow rate of the monomer until the average temperature in the reactor. The main variables affecting the duration of the reaction are: the temperature, initial amount of monomer, flow rate of the monomer and the addition of the initiator. The quality of the product is defined by the following parameters: conversion into polymer, particle size distribution, viscosity and solid content.

Modeling of Polyvinyl Acetate Polymerization Process for Control Purposses 3 3. Temperature model The model of the polymerization process of vinyl acetate was developed using gpoms modeling tool. The current model is based on previous work of Aller et al.6,7, where the monomer flow rate, initiator addition and the temperature were taken as inputs into the system. The temperature was not estimated by the model. In order to observe the changing of its profile as a result of the change of the initiator addition we took out the temperature as an input, but rather estimate it from the energy balance model. This enables us to controlling it by adding a sufficient amount of initiator. The most influential factors attributing to reaction rate are the decomposition of the initiator and the temperature in the reactor. The mass of the initiator in the reactor (I) is increased by the flow addition of initiator (I e ) and decreased by its thermal decomposition 9,10. To obtain the concentration of the initiator (I w ) in the aqueous phase we divide by the volume of water (V w ). di I e k d I w V w To account for higher rate of decomposition of KPS at higher temperature we use the following function K d 8.4 T Tstart 40 2 1.6 T Tstart 40 where K d is the initiator decomposition rate constant, T is the reactor temperature and Tstart is the temperature when an appreciable decomposition is observed. eaction heat capacity (K ) accounts for the heat capacity of the still present monomer and converted monomer (polymer) in the reactor. It is calculated by the specific heat capacities (Cp mon, Cp pol ) of the monomer (M), having into account its molecular weight (MW m ) and the converted monomer (M conv ). K MMWm Cpmon MconvMWmCppol (3) The flow through the condenser is estimated by the flow coming in and out of it (Q FC ), its heat capacities (Cp Cout ) and temperature of the reflux going through the condenser (T Cout, T Cin ). The λ is vaporization heat. (1) (2) Q ))) cond ( QFC( CpCout( TCout TCin QFC (4) 2 0.00281T 0.32738 408.178 (5) Cout T Cout We also take into account the heating through the heating. This is considered in the first part of the modeling the temperature profile when the is heated. In that first part we do not have heat looses and also there is no flow though the condenser, therefore we have the following equation for the energy balance: dt dk K T QmMWmCpmonTmon H rrpol H (6) Where on one side of the equation is the reaction heat capacity multiplied by the changes in the temperature in the reactor and vice versa. On the other side of the

4 T. Miteva equation is the monomer enthalpy, the reaction heat and the heating in the. The heating through the is given by: H.00018( T T ) K 0 (8) where the changes in the temperature in the is: dt 0.00018 K 4.18 *1000)( T T ) (8) ( The energy balance for the part of the model when no heating is needed anymore, does not include heating through the and takes into account heat losses (Q loss ) and flow through the condenser (Q cond ). The equation has the following form: dt dk K T QmMWmCpmonTmon H rrpol Qcond Qloss (9) Q.9502( T T ) (10) loss 0 ext Model parameters were taken from the literature 8 (Cp pol, Cp mon, Cp out ) or estimated based on real plant data (Tstart, Tstart ) so that a satisfactory agreement between the process and the model was obtained. The values of model parameters are given in Table I. Table I Variables estimated from the literature and expiriments Parameter Value Tstart [ K] 330 MW m [gmol -1 ] 86.09 Cp pol [kj kg -1 ] 1.77x10-2 Cp mon [kj kg -1 ] 1.77x10-2 Cp out [kj kg] 1.17 T ext [ K] 283.15 4. esults Model simulations were performed based on real plant data. The monomer flow rate was taken from the actual measurements, while the addition of initiator was calculated from the weight change of the initiator dozing tank (Figure 2). 100 90 80 70 60 50 40 30 0.004 0.0035 0.003 0.0025 0.002 0.0015 0.001 0.0005 T eactor C flow rate init testing mol/s 20 0 10000 20000 30000 40000 50000 60000 Fig 2Initiator addition and Temperature Profile in time, data taken from real plant With the presented model we are able to get relatively good estimates of the reactor temperature. From Figure 3 it can be seen that the estimated profile follows the 0

Modeling of Polyvinyl Acetate Polymerization Process for Control Purposses 5 dynamics of the real one. Not so good estimates are obtained at the beginning and at the end of the batch. At the beginning of the batch, the most influential temperature parameter is the heating of the reactor through the heating. So most probably, the corresponding model parameters still need to be estimated more precisely. The agreement of the model is also not good in the last part of the estimated profile. This part is when the whole amount of monomer is added in the reactor and a larger amount of initiator is put to finish the reaction. There are two possible reasons that the performance of the model in this part is not so good, i.e., (i) the heat losses were overestimated, (ii) the final addition of initiator was bigger than calculated. For the latter we have to take into account that the amount of initiator is obtained from the weight change of the dozing tank, so there might be some disparity of the final addition because of its larger amount. In the estimated temperature profile there is also a small delay in the increasing and decreasing of the temperature. The most probable reasons for that are the value of the initiator decomposition rate constant, heat losses and, most probably, the amount of initiator addition. The obtained model still needs fine tuning of parameter values and close inspection of the initiator addition. Figure 3 Temperature profile and estimated profile Process outputs were also estimated with the model. Simulations were done for four batches. Table 1 shows the validation of the model against real plant data. For the viscosity and the solids content, laboratory analyses were performed at the end of the reaction. These measurements are normally performed after mixing several batches and therefore no individual data is available. Expected output results for the conversion are over than 99%. The range for the solids content is 44 46, the viscosity has to be over than 34000. Table II Output Variables real data and estimated data Batch No. Conversion (%) Solids Content (%) Viscosity (Brookfield 200C, 50rpm), (mpas) eal Modeled eal Modeled eal 1192 99.95 99.96 46.4 47.2 37520 36166 1214 99.90 99.97 45.9 46.81 31200 35880 1253 99.13 99.95 46.9 47.70 34000 36876 1256 99.37 99.86 46.4 47.16 30160 35261 Modeled data From the table we see that some of the model outputs are overestimated, but are still in the quality range of the product. Also for two of the batches (1214 and 1256) the

6 T. Miteva viscosity, as measured in the plant, is less than it should be. This output variable is in its range in the modeled process. 5. Conclusion A model for polymerization process of vinyl acetate has been developed. The model extends the work of Aller et al. 7 in terms of energy balance. The improved model has two inputs flow rate of monomer and flow rate of initiator addition. The outputs of the presented model are four particle size diameter, solids content, viscosity and conversion. Having the temperature as an additional state variable we are able to estimate more precisely the sufficient amount of initiator which should be added into the system every time the temperature drops off. The model has been validated on a new set of real data taken from the polymerization plant. The work continues with fine tuning of model parameters, developing optimal control strategies for monomer and initiator addition, and implementing them in chemical factory in Sežana, Slovenia. 6. Acknowledgment The support of the European Commission in the context of the 6th Framework Programme (PISM, Contract No. MTN-CT-2004-512233) is greatly acknowledged. eferences 1. J. T. O Donnell and. B. Mesrobian (1958). Vinyl acetate emulsion polymerization. Journal of Polymer Science, Vol. 28, 171-177. 2. S. Dunn and P. A. Taylor (1965). The polymerization of vinyl acetate in aqueous solution initiated by potassium persulphate at 60ºC. Makromol., 83, 207-219. 3. M. B. Urquiola, V. L. Dimonie, E. D. Sudol, M. S. El-Aasser, (1992) Journal of Polymer Science: Part A: Polymer Chemistry, 30, 12, 2631-2644 4. M. B. Urquiola, V. L. Dimonie, E. D. Sudol, M. S. El-Aasser, (1992) Journal of Polymer Science: Part A: Polymer Chemistry, 30, 12, 2619-2629. 5. M. B. Urquiola, V. L. Dimonie, E. D. Sudol, M. S. El-Aasser, (1992) Journal of Polymer Science: Part A: Polymer Chemistry, 31, 6, 1403-1415. 6. Aller, Fernando, Kukanja, D., Modeling of the semi-batch vinyl acetate emulsion polymerization in an industrial reacor, CHISA 2006, 27-31 August 2006, Prague Czech epublic. 7. Aller, F., Kandare, G., Blázquez Quintana, L. F., Kukanja, D., Jovan, V. and Georgiadis, M. (2007). Model-based optimal control of the production of polyvinyl acetate. European Congress of Chemical Engineering (ECCE-6), Copenhagen, 16-20 September, 2007 8. K.C. Berger and G. Meyerhoff, Polymer handbook, 3 th edition, Wiley, New York (1989) 9. C.M Gilmore, G.W. Poehlein and F.J. Schork, (1993) Journal of Applied Polymer Science, 48, 1449-1460. 10. C.M Gilmore, G.W. Poehlein and F.J. Schork, (1993) Journal of Applied Polymer Science, 48, 1461-1473.