Fault Tolerant MPC for Fed-Batch Penicillin Production

Abstract

Biopharmaceutical manufacturing is an inherently complex operation with industrial-scale control strategies often failing to accurately account for faults and product variation that can result in significant economic losses. In response, this work presents a fault-tolerant Model Predictive Control (MPC) strategy which solves a modified version of a Quadratic Programming (QP) problem, using a data-driven adaptive Multivariate Partial Least Square (MPLS) model. The proposed technique is applied to an industrial simulation of a fed-batch process and found to improve operations significantly.