(712a) Economic Nonlinear Model Predictive Control of Continuous Pharmaceuticals Manufacturing Processes
In previous work the problem of offline economic open-loop optimal control was investigated on toy problems  as well as on more realistic plant-wide models . The problem formulation involves a hybrid (discrete-continuous) dynamical system, where switching between on/off-spec production modes occurs at discrete times, which are themselves optimization variables. These modes' order is predetermined, according to the turnpike property of optimal dynamic systems, allowing to perform correct sensitivity analysis, to be used in gradient based optimization algorithms. The dynamic optimization approach maximizes the accumulated on-spec production directly over the entire time horizon. In this sense the process being controlled has properties of both continuous and batch processes.
When considering online control, previous work has focused mainly on controlling a predetermined steady state operation, using traditional PI controllers  or MPC , without considering the economics of the overall production. Here, on the contrary, we focus on real-time control strategies that optimize an economic objective function. Some of the challenges this approach is facing are: 1. The dynamic optimization involves nonsmooth and often hybrid behavior, which results in high computational cost of the sensitivity analysis. 2. Solving the optimal control profile for shorter times than the actual campaign time may result in different (sub-optimal) solutions.
In this contribution we study the implementation of non-linear model predictive control (NMPC) schemes to operate such a plant. An economic-NMPC approach is investigated for online optimal operation of the manufacturing process. We compare the results of a bench-scale ideal receding horizon economic-NMPC approach to a hierarchical approach, where the control is conducted in two layers: 1. On the top we have a real-time dynamic optimization layer that considers the entire time horizon (campaign time) 2. Below we apply NMPC controllers using simplified models and shorter horizons. We also make comparison to an open-loop optimal control. Choosing the control variables and dynamic set-points will also be discussed.
1. S. Mascia et al., "End-to-End Continuous Manufacturing of Pharmaceuticals: Integrated Synthesis, Purication, and Final Dosage Formation", Angew Chem. Int. Ed. Engl., 2013, 52(47),12359-63.
2. A. Sahlodin and P.I. Barton, "Optimal Campaign Continuous Manufacturing", Ind. Eng. Chem. Res., 2015, 54(45), 11344-59.
3. M. Shoham Patrascu and P.I. Barton, "Modeling and Dynamic Optimization of an End-to-End Continuous Pharmaceuticals Manufacturing Plant", in preparation.
4. R. Lakerveld et al., "The Application of an Automated Control Strategy for an Integrated Continuous Pharmaceutical Pilot Plant.", Org. Process Res. Dev., 2015, 19, 1088-1100.
5. A. Mesbah et al., "Plant-wide Model Predictive Control for a Continuos Pharmaceutical Process", In American Control Conference (ACC), 2015, pp. 4301-4307.