UC Berkeley

This paper considers the model predictive control (MPC) of critical quality attributes (CQAs) of products in an end-to-end continuous pharmaceutical manufacturing pilot plant, which was designed and constructed at the Novartis-MIT Center for Continuous Manufacturing. Feedback control is crucial for achieving the stringent regulatory requirements on CQAs of pharmaceutical products in the presence of process uncertainties and disturbances. To this end, a key challenge arises from complex plant-wide dynamics of the integrated process units in a continuous pharmaceutical process, that is, dynamical interactions between several process units. This paper presents two plant-wide MPC designs for the end-to-end continuous pharmaceutical manufacturing pilot plant using the quadratic dynamic matrix control algorithm. The plant-wide MPC designs are based on different modeling approaches - subspace identification and linearization of nonlinear differential-algebraic equations that yield, respectively, linear low-dimensional and high-dimensional state-space models for the plant-wide dynamics. The closed-loop performance of the plant-wide MPC designs is evaluated using a nonlinear plant simulator equipped with a stabilizing control layer. The closed-loop simulation results demonstrate that the plant-wide MPC systems can facilitate effective regulation of CQAs and flexible process operation in the presence of uncertainties in reaction kinetics, persistent drifts in efficiency of filtration units, temporary disturbances in purity of intermediate compounds, and set point changes. The plant-wide MPC allows for incorporating quality-by-design considerations into the control problem through input and output constraints to ensure regulatory compliant process operation.