Using a Continuous MPC Algorithm to Control An Unsteady Batch Fermentation Process

Classical MPC has been leveraged to improve control on a nonlinear unsteady-state system, batch alcohol fermentation. The controller uses a hybrid model with fundamental reaction equations along with empirical process data-based modeling. More than thirty lines of batch quality control have been successfully commissioned. These applications include property estimators, optimal batch trajectory management and classical sense and respond MPC with lab feedback within control algorithms that were designed to use dynamics on stable processes.

The algorithmic stretch occurs because of the instability. With classical MPC algorithms, step tests produce a stable response within an operating condition. However, in an unstable batch fermentation, by the time one observes a process response, a new operating condition already has been established. The solution is to leverage hybrid modeling using known fundamental fermentation equations, tuned through process historic data and dynamic testing to match the process performance. This hybrid modeling system runs against the process and targets a dynamically optimal batch quality trajectory. Periodic lab analyses of the fermenter contents provide feedback. Yield improvements have been documented.