APPLICATION OF NONLINEAR MODEL PREDICTIVE CONTROL BASED ON DIFFERENT MODELS TO BATCH POLYMERIZATION REACTOR Sevil Cetinkaya a,* , Duygu Anaklı a , Zehra Zeybek b , Hale Hapoğlu b , Mustafa Alpbaz b a Cumhuriyet University, Department of Chemical Engineering, 58140, Sivas, Turkey b Ankara University,Department of Chemical Engineering, 06100 Tandogan, Ankara, Turkey * e-mail: cetinkaya.sevil@gmail.com Phone: +905055671933, Fax: +903462191179 ABSTRACT In the present work, at the previously determined optimal conditions, to control batch polymerization reactor, Linear Generalized Predictive Control (LGPC) and Nonlinear Generalized Predictive Control (NLGPC) algorithm were utilized. Several system models were applied to the control algorithms. The effect of different optimal conditions has been examined on monomer conversion, average viscosity molecular weight and chain length. At the same operating conditions of LGPC and NLGPC temperature control was used for comparison. According to the experimental results, the performance of NLGPC was obtained well than LGPC control method. In addition, the results denoted that the NLGPC control performances depend on different models and the optimum conditions. Keywords: Styrene polymerization, Generalized Predictive Control, Nonlinear Model Predictive Control 1. INTRODUCTION Control of polymerization reactors is often difficult and sensors to provide on-line measurement of polymer properties are generally not available (Altınten, Erdogan, Hapoglu, and Alpbaz 2003; Altınten, Erdogan, Hapoglu, Alıev, and Alpbaz 2006; Cetinkaya 1996). The most significant task for a polymerization reactor control strategy is to maintain the major design because of having complex and nonlinear reaction and operational variables like product quality is also important to preserve smooth and stable operation. Physical, chemical and mechanical properties of polymers are generally closely related with their molecular weights. But the weight of all polymer molecules within a polymer sample is not equal to each other. For this reason, the molecular weight of polymers that were determined in any way shows average number not the exact values. The full molecular weight distribution (MWD) of a specific polymer and ratio of moments of this distribution, such as the number average ( n M ) or the weight average ( w M ) molecular weight, indicate the mechanical properties of the polymer. Initial initiator concentration and temperature are the primary control ways to influence the molecular weight of a polymer produced in free radical polymerization (Barner-Kowollik and Davis 2001). The temperature change has been observed that has great influence on the kinetics of polymerization process, and physical properties and quality of produced polymer. The main objective of the temperature control of polymerization reactor is to remove a great amount of heat from the exothermic reaction to achieve the desired number average chain and a desired conversion in a minimum time (Yuce, Hasaltun, Erdogan, and Alpbaz 1999). Then, reactor temperature should be controlled effectively to satisfy the desired polymer quality. Various control methods have been applied both theoretically and experimentally to the systems at constant and changing set points (Zeybek, Cetinkaya, Hapoglu, and Alpbaz 2006; Seki, Ogawa, Ooyama, Akamatu, Ohshima, and Yang 2001). Lewis, Nguyen, and Cohen (2007), highlighted to the effect of initiator amount on the radical polymerization. In free-radical polymerization, the reaction temperature and both the initiator and the chain-transfer agent’s concentrations are usually chosen as controlled variables. These variables can also affect the rate of polymerization and the molecular weight of the polymer (Ponnuswamy, Shah, and Kiparissides 1987). Generalized Predictive Control (GPC) algorithm is commonly used in polymerization reactors (Ozkan, Hapoglu, and Alpbaz 1998; Yuce (Çetinkaya) 2001). But, NLGPC algorithm was rarely practiced with different models and at constant temperature in a batch polymerization reactor. Zeybek, Cetinkaya, Hapoglu, and Alpbaz (2006) developed the generalized delta rule 148