(103c) Analysis of Foaming in Batch Fermentation Processes Using Observation-Wise and Batch-Wise Unfolding Partial Least Square (PLS) Approaches

This work presents an essential application of multivariate data analysis to fermentation process monitoring and quality enhancement, especially the prediction of the occurrence of foaming in fermenter operation. Foaming is a phenomenon that frequently occurs inside fermenters and reduces their effective working volume, resulting in a decrease in product yields. This study performs a constructive quantitative analysis of predicting foaming occurrence using a technique commonly used in chemometrics called partial least square (PLS), extended to Multiway PLS (MPLS) for three-dimensional data matrices. At the same time, we examine the relationships and correlations between the product quality measurements and the independent process variables. Consequently, we try to analyze those relations to the foaming problems in the fermenter. In addition, we also compare and contrast two unfolding methods, Observation-Wise Unfolding (OWU) and Batch-Wise Unfolding (BWU). To our knowledge, no published study has demonstrated the applications of multivariate data analysis in the foaming prediction of the fermentation processes and the comparison between the two aforementioned unfolding techniques. Concerning the OWU method, we use two approaches of MPLS, Kernel Partial Least Square (KPLS) and Dynamic Partial Least Square (DPLS), where we consider the exhaust differential pressure as the dependent variable. With respect to the BWU method, we refer to the product quality variables as dependent variables and apply PLS to correlate the dependent variables to process variables. The OWU KPLS/DPLS approaches show that the exhausted differential pressure positively correlates with volume and pH while negatively correlates with dissolved oxygen. In the BWU approach using PLS, we conclude that foaming has a negative impact on the quality variables. We recommend the BWU approach when analyzing the differences among batches and monitoring the product quality, and we prefer the OWU approach when the dependent variable is time-dependent. Thus, in this study, we recommend the OWU approach for the comprehensive analysis of foaming.