(528c) A Predictive Control Approach for Processes with Multiscale Objectives

For many chemical and biological processes of industrial interest, performance is measured both with respect to product yield and quality. The elusive latter term is usually dependent upon enforcing the product microstructure within strict limits. Multiscale models are traditionally used to quantify process evolution across all relevant length scales and characterize product behavior within the current computational limitations. Even such models however, pose significant challenges both from an analysis and control point of view [1]. Such difficulties are attributed in part to the unavailability of closed form models to describe the process evolution at molecular-level detail and their computationally intensive nature that prevents their real time implementation. An industrially relevant example is thin film deposition processes widely used by the microelectronics and solar energy industries (such as the production of photovoltaic systems). Due to the complex process dynamics and the strict quality requirements, a significant amount of research has focused on the design of feedback control structures. To circumvent the mentioned limitations, one of the proposed approaches identifies stochastic partial differential equation models to design the controller [2]. This approach however assumes specific structure to the nonlinear stochastic terms. Another approach relies on the off-line and subsequently the on-line identification of bilinear models for the process, which are then used for the controller synthesis [3]. The objective of the present work is to develop a computationally tractable predictive controller which explicitly takes into account the nonlinear nature of the underlying process dynamics. Specifically, we initially develop a computationally tractable online system-identification scheme based on subspace-system identification methods to address the above mentioned problems [4,5]. The proposed method explicitly accounts for the presence of stochastic terms in the process and the issue of noise in the measurements. The developed model is then employed with a model predictive framework to drive the process at the desired process objective. By employing topology concepts that account for the nonlinear nature of the underlying dynamics, the predictive controller is capable to efficiently identify optimal manipulated variable profiles. The parallel processor computer architectures that are currently becoming commonplace are naturally exploited to further accelerate the convergence of the controller to the solution.

REFERENCES: [1] P. D Christofides, A Armaou , ?Control of Multiscale and Distributed process systems-preface,? Computers &Chemical Engineering., 29,687-688, 2005. [2] Y Lou, P. D Christofides, ?Nonlinear Feedback Control of Surface Roughness Using a Stochastic PDE: Design and Application to a Sputtering Process,? Industrial & Engineering Chemistry Research., 45, 7177-7189, 2006. [3] A. Varshney, A Armaou, ?Multiscale Optimization using hybrid PDE /KMC process systems with applications to thin film growth?, Chemical Engineering Science., 60, 6780-6794, 2005. [4] W. Favoreel, B. D Moor and P. V. Overschee, ?Subspace state space system identification for industrial processes,? Journal of Process Control., 10, 149-155, 2000. [5] A Armaou, ?Output feedback control of dissipative distributed process via microscopic simulations?, Computers &Chemical Engineering., 29, 771-782, 2005.