Track: Modeling and Simulation

Abstract

This work was directed to the application of advanced machine learning and process control tools in an industrial penicillin production bioreactor. The developed algorithms were the Principal Component Analysis (PCA) method, for reducing data dimensionality, and Radial Basis Function Neural Networks (RBFNN), directed to create a multivariable interpolation model to predict the bioreactor temperature. Regarding the process control phase, the Model Predictive Control (MPC) method was used, which aims to calculate a sequence of control actions capable of approximating the output variables with their respective set points in an optimized way. In this paper, information from a real industrial process of penicillin production was analyzed, dealing with a database composed of 27 variables to be treated. The gaussian radial base function gave the neural network model a better adaptation to the system. Finally, the control actions calculated by the MPC controller resulted in the maintenance of the controlled variables in its respective set points in a gradual manner, without impacting the stability of the system.