An Investigation of Multivariate Process Control for Biomanufacturing

Abstract

Biologics manufacturing is primarily managed through single loop univariate or cascaded controls - technology that has not fundamentally changed in decades. Outside of biomanufacturing, process control technologies have advanced to include multivariate and predictive control. This project examines the feasibility of developing a generalized multivariate control scheme proof of concept to link several individual loops by evaluating impact on control quality. A multivariate simulation environment was created to model the reactions of different control schemes on the critical quality attributes under investigation. This study reveals that model predictive control can be used on bioreactor control in this simulated environment; however, the results do not match the PID loops as closely as expected. While the multivariate model predictive control scheme shows a shift in mean difference off set point than the traditional control scheme in this purely simulation based experiment, it may be sufficient if additional benefits, such as better insight into more significant critical quality attributes, can be ascertained. Several future uses of this technology are hypothesized and, with additional effort, can be virtually tested given the baseline simulations from this project. Future testing can be implemented using this framework environment that can test the hypotheses that may lead to tighter control of quality attributes, increase in high quality titer, or lower waste of materials.