(610g) Quality Relevant Model Predictive Control for Batch Processes Via Subspace Identification

Batch processes play an important role in chemical, pharmaceuticals, bio-chemicals and materials industries for the production of low-volume but high-value-added products. The nonlinear and time-varying dynamics over a wide range of operating conditions and absence of equilibrium points preclude the direct application of control strategies designed for continuous systems. The primary control objective in batch processes is to reach a specified product quality by batch termination to consistently produce products of desired quality.

The existing batch process control strategies can be divided into trajectory tracking and inferential quality approaches [1, 2]. In trajectory tracking control, it is not guaranteed to meet to the desired quality even with perfect tracking if there is significant variation in the initial conditions from batch to batch in that disturbances encountered during the new batch could alter the relationship between the product quality and the trajectories of the process variables. In parallel, the data-driven inferential quality control is developed through multivariate statistical process control, such as partial least squares (PLS) regression beginning with batch-wise unfolding of multiway batch data. The multiway analysis based inferential model calls for the entire batch trajectory to predict the quality of the batch. However, measurements are only available up to the current sampling instant during the batch operation, and the future data required to predict the final quality is not measured yet, which is treated as a missing data problem [3].  In reality, the problem is not one of missing data but rather one of dynamic modeling. Therefore, a data-driven dynamic model is recently developed for quality control in batch processes, integrated with multiway PLS [4]. Nevertheless, multiway analysis based inferential model includes process measurement trajectories that only indirectly predict product quality, because it is recognized that the final product quality only depends on the final state of the process.

Motivated by above considerations, a subspace identification based quality relevant control strategy is developed for batch processes in this work. To this end, canonical variate analysis (CVA) method is employed to identify a dynamic state space model for the batch process [5]. Subsequently, an inferential model is constructed to predict the final product quality using the identified final state. A model predictive controller is then designed to achieve desired final product quality. The proposed approach is applied to a penicillin fermentation process and the simulation results indicate that the superior performance of the proposed controller over the traditional trajectory tracking controller.

[1] Cruickshank, S. M., Daugulis, A. J., & McLellan, P. J. (2000). Dynamic modeling and optimal fed-batch feeding strategies for a two-phase partitioning bioreactor. Biotechnology and bioengineering, 67(2), 224-233.

[2] Golshan, M., MacGregor, J. F., Bruwer, M. J., & Mhaskar, P. (2010). Latent Variable Model Predictive Control (LV-MPC) for trajectory tracking in batch processes. Journal of Process Control, 20(4), 538-550.

[3] Flores-Cerrillo, J., & MacGregor, J. F. (2004). Control of batch product quality by trajectory manipulation using latent variable models. Journal of Process Control, 14(5), 539-553.

[4] Aumi, S., Corbett, B., Clarke‐Pringle, T., & Mhaskar, P. (2013). Data‐driven model predictive quality control of batch processes. AIChE Journal, 59(8), 2852-2861.

[5] Larimore, W. E. (2013, June). Identification of nonlinear parameter-varying systems via canonical variate analysis. In American Control Conference (ACC), 2013 (pp. 2247-2262). IEEE.