(591e) Implementation of Model-Predictive Control in Continuous Crystallization

Automation and process control are essential when operating a continuous (flow) process to minimize waste and start up time. Typical operating protocols involve implementing the steady state inputs such as feed concentration and reactor jacket temperature immediately at start-up, then divert the out-of-spec material until the process achieves steady-state. From that point the product is collected. In this study, we implement a model-predictive controller to start up a continuous crystallizer optimally, achieve and maintain steady-state conditions faster than open-loop recipe driven operations. Reactor temperature and solution concentration are the measured variables. The solution concentration is measured using a ReactIR probe calibrated using absorbance as a function of solution concentration. The feed concentration of the inlet stream and the reactor jacket temperature are the manipulated variables. This presentation will compare open-loop and closed-loop controlled concentration profiles in a continuous crystallization for a small molecule API and for a chemical intermediate at Biogen. The closed-loop controller results in steady-state achieved in approximately three residence times, whereas it takes five residence times for the open-loop case to reach steady state. Crystal habit and size are also monitored and show size trends as a function of supersaturation.