Masoud Golshan, Chemical Engineering, PhD Candidate
25 March 2010 at 10:30
Location: JHE 326H
Batch processes exhibit a number of characteristics that lead to interesting control problems. They are finite duration processes in which the objective is to achieve a desired product by the very end of the batch. Moreover, they are nonlinear and time varying in that the gains and dynamics often vary continuously throughout the duration of the batch. There are basically two levels of control for batch processes. The higher level control is the control of final product quality at the end of the batch. The lower level control is the set-point tracking of certain process variable trajectories that have the most effect on the final product quality. Model predictive control (MPC) is a class of advanced controllers that can be used for the trajectory tracking control. It incorporates future prediction of process outputs, by utilizing the process model, predicts the effect of disturbances on the outputs, and takes corrective actions well in advance. Although batch processes are t!
he heart of many industries such as pharmaceuticals and specialty chemicals, most of the research on MPC is concentrated on continuous processes and there is no dedicated MPC for batch processes. The main bottleneck in the application of MPC for batch processes is the process model. The more accurate the process model, the better performance obtained from the MPC algorithm. Inaccuracies caused by simplifying assumptions in mechanistic models together with costly procedure of developing such models makes empirical models more attractive. During my Ph.D. study, we have developed a dedicated Model Predictive Control (MPC) for batch processes which uses an empirical model called Latent Variable Model (LVM) for the prediction step. This algorithm is called Latent Variable MPC (LV-MPC). The Principal Component Analysis (PCA) is the selected LVM to be used in the proposed algorithm.
Return to list of all seminars