Subspace identification and predictive control of batch particulate processes

Abstract

This paper addresses the problem of subspace identification based modeling and predictive control of batch particulate process with an application to crystal size distribution (CSD) control in a batch crystallizer. To this end, a subspace identification technique is first adapted to identify a linear time invariant model for batch particulate processes. The estimated model is then deployed in a linear model predictive control (MPC) formulation to achieve a particle size distribution with desired characteristics subject to both manipulated input and product quality constraints. The proposed approach is implemented on a seeded batch crystallizer process and compared with an open loop policy as well as a PI controller based trajectory tracking policy. The proposed MPC is shown to achieve 27% and 30% improvements, respectively.