DEEP KOOPMAN NEURAL NETWORKS FOR NONLINEAR PROCESS MONITORING IN STOCHASTIC PRODUCTION SYSTEMS

Abstract

Stochastic production systems (SPS) play a pivotal role in industries such as fermentation, pharmaceuticals, and composite material production, where stringent quality constraints are paramount. To ensure product quality in such systems, effective process monitoring is imperative. However, SPS presents significant challenges due to its inherent stochasticity and measurement uncertainties, stemming from sensitivity to exogenous factors and the lack of accurate in-situ measurements. This paper explores the landscape of SPS process monitoring methods, highlighting their limitations and proposing a novel approach leveraging recurrent neural networks (RNNs), particularly Long Short-Term Memory (LSTM) networks