(443a) Plant-Wide Model Predictive Control of a Continuous Pharmaceutical Manufacturing Process

The pharmaceutical industry is undergoing revolutionary changes to respond to increased demands for more efficient and cost-effective processes through the application of continuous manufacturing [1,2]. The enhanced process actuation in continuous manufacturing combined with the use of the state-of-the-art process analytical tools for process monitoring have created vast opportunities for online process control in the pharmaceutical industry. Aligned with quality-by-design (QbD) concepts [3], online control of continuous pharmaceutical manufacturing processes is intended to enable achieving the desired critical quality attributes (CQAs) of the end products consistently in the presence of various process disturbances and uncertainties.    

Advanced control of isolated unit operations (e.g., crystallizers, granulation units, compaction units, etc.) in pharmaceutical processes has been extensively investigated in the literature. Effective control of continuous pharmaceutical manufacturing processes however necessitates exploring the dynamics of several interconnected unit operations with recycle, bypass, and heat streams as a whole. The interactions between various unit operations through material and energy streams significantly complicate the plant-wide process dynamics, which will lead to poor performance of the decentralized unit operation level control systems [4]. Hence, a plant-wide control strategy is required for the integrated continuous manufacturing process to actively maintain the critical process parameters (CPPs) within their design spaces to realize the stringent regulatory requirements on the CQAs of the end products.

This work investigates plant-wide model predictive control of an end-to-end continuous manufacturing pharmaceutical process, which is inspired by a pilot plant built within the Novartis-MIT Center for Continuous Manufacturing. The process manufactures a pharmaceutical product from start (synthesis of intermediates) to finish (coating of drug product in tablet form) in a fully continuous manner. A nonlinear plant-wide dynamic model of the pilot plant [5] is utilized to simulate the dynamics of the real process. A stabilizing control layer is incorporated into the plant simulator to ensure effective inventory control throughout the plant. A linear time-invariant (LTI) state-space model of the plant is identified by using the plant simulator to generate input-output data around the desired steady-state of the process and, subsequently, applying the N4SID subspace identification algorithm [6]. The LTI model is used to characterize the finite step response (FSR) dynamics of the process with respect to manipulated variables (CPPs) and measured disturbance variables of the plant. The FSR models constitute the underlying prediction model of the quadratic dynamic matrix control (QDMC) algorithm [7] designed for plant-wide control of the continuous pharmaceutical manufacturing process.

The performance of the model predictive controller is assessed in closed-loop operation with the nonlinear plant simulator. The control objective is to sustain the CQAs (solvent content, total level of impurities, and dosage of API) of the final product in their admissible ranges, while achieving a desired tablet production rate. Three scenarios are considered to examine the closed-loop performance of the controller:

1- Uncertainties in reaction kinetics for synthesis of the intermediate as well as the API compounds.

2- Persistent disturbances arising from performance degradation of filtration units.   

3- Temporary disturbances due to variations in the purity level of raw materials.

The objective of this talk is twofold: first to demonstrate the design of a plant-wide model predictive controller for a continuous pharmaceutical manufacturing process, and second to evaluate the performance of the model predictive controller with that of the plant-wide regulatory control strategy presented in [8]. 

[1] H. Leuenberger. New trends in the production of pharmaceutical granules: Batch versus continuous processing. European Journal of Pharmaceutics and Biopharmaceutics, 52:289-296, 2001.            

[2] P. McKenzie, S. Kiang, J. Tom, A. E. Rubin, and M. Futran. Can pharmaceutical process development become high tech? AIChE Journal, 52:3990-3994, 2006.

[3] L. X. Yu. Pharmaceutical quality by design: Product and process development, understanding, and control. Pharmaceutical Research, 25:781-791, 2008.

[4] S. Xu and J. Bao. Distributed control of plantwide chemical processes. Journal of Process Control. 19:1671–1687, 2009.

[5] B. Benyahia, R. Lakerveld, and P. I. Barton. A plant-wide dynamic model of a continuous pharmaceutical process. Industrial and Engineering Chemistry Research. 50:10083-10092, 2011.

[6] L. Ljung. System identification: Theory for the user. 2^ndEdition. Prentice Hall PRC, New Jersey, 1999.

[7] C. E. Garcia and A. M. Morshedi. Quadratic programming solution of dynamic matrix control (QDMC). Chemical Engineering Communications. 46:73-87, 1986.   

[8] R. Lakerveld, B. Benyahia, R. D. Braatz, and P. I. Barton. A plant-wide control strategy for a continuous pharmaceutical pilot plant. Accepted AIChE Journal. 2013.