Process‐aware data‐driven modelling and model predictive control of bioreactor for the production of monoclonal antibodies

Abstract This manuscript addresses the problem of controlling a bioreactor to maximize production desired product while respecting constraints imposed by nature bio‐process. The approach is demonstrated first building data‐driven model and then formulating predictive controller (MPC) with results illustrated implementing detailed monoclonal antibody (the test bed) created Sartorius Inc. In particular, recently developed modelling using an adaptation subspace identification techniques utilized that enables incorporation known physical relationships in development (constrained identification), making process aware. resultant implementation demonstrates significant improvement compared existing proportional integral (PI) strategy used production. Simulation also demonstrate superiority process‐aware or constrained MPC traditional MPC. Finally, robustness design via data from bed different set parameters, thus showing ability maintain good performance event changes such as cell line feed characteristics.