Exploratory Analysis of Bioprocesses Using Artificial Neural Network‐Based Methods

TLDR

Industrial bioprocesses are often not comprehensively investigated, creating a need for efficient analysis methods. This study validates the proposed procedure using simulated bioprocess data. The approach employs artificial neural networks with mass‑balance equations, starting from a minimal model of essential variables and iteratively adding others that pass a performance‑enhancement test. The resulting numerical model captures key process relationships, enabling improved set points, profiles, state estimation, and control, and has already been applied successfully in industrial settings.

Abstract

Abstract A process data driven procedure has been developed that allows a universal time‐efficient bioprocess analysis. The procedure is particularly suited for industrial production processes which have not yet been comprehensively investigated. It makes use of artificial neural networks in combination with mass balance equations to represent the process dynamics on a commercial workstation. The essential concept behind the procedure is to start with the already available knowledge formulated by a very simple process representation which includes only those variables that are firmly known to be essential. Then, stepwise, additional variables are added to the basic representation after they passed a test procedure in which they proved to enhance the model's performance. The result of the procedure is a numerical representation of the important process relationships that immediately allows to determine improved set points and/or profiles for the manipulated variables with respect to process performance. It may be used to improve state estimation and control. The procedure has already been tested in industrial applications. In this paper, a validation of the procedure with simulated bioprocess data is presented.