(432g) Nonlinear Model Predictive Control for a Continuous Pharmaceutical Manufacturing System: A Comparison of Control Strategies for a Thin-Film Formation Process

Continuous manufacturing is receiving increasing attention in the pharmaceutical industry to reduce time-to-market and production costs while enhancing product quality [1,2]. In addition to improved flexibility, reliability, and efficiency of the production process, continuous manufacturing facilitates the use of increased process understanding for online process control [3]. This operational mode can lead to consistently high-quality products as well as reduced waste generation and energy consumption.     

To address the manufacturing challenges of solid-form drugs (e.g., tablets and capsules), thin films that dissolve quickly have been developed as an oral drug delivery system in recent years [4,5]. Thin-film manufacturing is based on solutions and therefore alleviates the solids-handling problems. Thin-film dosage forms are especially advantageous when the Active Pharmaceutical Ingredient (API) cannot be dispersed well in a solid form or the solids handling reduces the API yield [6], but the use of dissolving oral films is limited to APIs with fast metabolic uptake rates due to the rapid disintegration of thin films.

A process of manufacturing pharmaceutical tablets from thin films has been developed at the Novartis-MIT Center for Continuous Manufacturing. The continuous process for making thin-film tablets consists of four steps: preparation of the formulation solution, casting the solution as a thin layer that is dried to produce the thin film, folding of the dried thin film, and compaction of the folded thin film to form tablets. This process combines the merits of thin-film manufacturing in terms of minimal solids handling and fast drying times with the wider applicability of tablets for effective drug delivery.

This talk designs and compares model-based control strategies for the control of properties along the machine direction of a continuous thin-film dryer, which comprises the second step of the thin-film tablet manufacturing process (see [7] for a discussion of machine direction vs. cross-directional control problems). The film is cast and dried in the thin-film dryer to remove solvents of the drug formulation solution. The mechanical characteristics and adhesion properties of thin films rely heavily on solvent remaining in the film at the end of the drying process, while the quality of produced tablets is affected by the film thickness. Controlling the solvent content and thickness of thin films in the dryer is important in the overall process of thin-film tablet formation.

The control objective for the thin-film dryer is to regulate the film properties (composition and thickness) as the film moves through the dryer. An interesting characteristic of this control problem is that the dynamics of composition evolution are strongly nonlinear. In addition, only a limited number of in-situ sensors are used to measure the uniformity of film properties (especially film thickness) for online control applications.

A nonlinear model predictive controller is developed for optimal operation of the thin-film dryer. The use of model predictive control has long been considered for the control of sheet and film coating processes (e.g., see [8,9,10]). What distinguishes this work is the complex nonlinear distributed dynamics of composition evolution and the nonlinear film shrinkage throughout the drying process that motivate the consideration of nonlinear state estimation and control techniques. Due to the change in film properties as the thin film moves along the dryer, there is a one-to-one correspondence between the drying time and the position of the film in the dryer. This implies that the control problem should be formulated with respect to a spatial domain and therefore the resulting operating policies correspond to the position of the thin film in the dryer at each time.

The cornerstone of the nonlinear model predictive controller is a dynamic optimizer that computes optimal operating policies in an online manner. In contrast to classical model predictive control (in which a nonlinear model can only be used for process simulation and the quadratic program is always based on a linear model), dynamic optimization enables using a nonlinear process model for both simulating the process dynamics over a prediction horizon and, subsequently, computing the optimal control inputs. An unscented Kalman filter [11] is designed to facilitate closed-loop implementation of the dynamic optimizer and to estimate unmeasured process variables (e.g., film thickness). The performance of the nonlinear model-based controller is compared to that of a more classical control strategy using two control scenarios. The simulation results indicate that the classical control system leads to a process operation comparable to that of the nonlinear model predictive controller. These results are explained by observing that the solvent concentration in the film is critical to most of its quality attributes, so that effective tracking of an admissible solvent concentration setpoint using a classical control system enables achieving adequate quality attributes of the thin film. Compared to the classical control system, the nonlinear model predictive controller does improve the energy efficiency of the thin-film dryer.

[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] N. Kockmann, M. Gottsponer, B. Zimmermann, and D. M. Roberge. Enabling continuous flow chemistry in microstructured devices for pharmaceutical and fine chemical production. Chemistry A European Journal, 14:7470-7477, 2008.

[4] A. Dinge and M. Nagarsenker. Formulation and evaluation of fast dissolving films for delivery of triclosan to the oral cavity. AAPS PharmSciTech, 9:349-356, 2008.

[5] J. J. Hirani, D. A. Rathod, and K. R. Vadalia. Orally disintegrating tablets: A review. Tropical Journal of Pharmaceutical Research, 8:161-172, 2009.

[6] S. Kunte and P. Tandale. Fast dissolving strips: A novel approach for the delivery of verapamil. Journal of Pharmacy and Bioallied Sciences, 2:325-328, 2010.

[7] J. G. VanAntwerp, A. P. Featherstone, R. D. Braatz, and B. A. Ogunnaike. Cross-directional control of sheet and film processes. Automatica, 43:191-211, 2007.

[8] J. C. Campbell and J. B. Rawlings. Predictive control of sheet- and film-forming processes. AIChE Journal, 44:1713-1723, 1998.

[9] J. G. VanAntwerp and R. D. Braatz. Fast model predictive control of sheet and film processes. IEEE Transactions on Control Systems Technology, 8:408-417, 2000.

[10] A. Halouskova, M. Karny, and I. Nagy. Adaptive cross-directional control of paper basis weight. Automatica, 29:425-429, 1993.

[11] S. J. Julier and J. K. Uhlmann. A new extension of the Kalman filter to nonlinear systems. In Proceedings of the 11th International Symposium on Aerospace/Defense Sensing, Simulation and Controls, pages 182-193, Orlando, USA, 1997.