Monitoring Process Phase Sequences and Durations in Industrial Batch Processes by a Hybrid Unsupervised Learning Strategy

Industrial batch processes, such as those in the pharmaceutical industry, are characterized by complex system behaviors, due to the involvement of several chemical reactions in various time critical process phases. Monitoring such processes involves several critical aspects: identifying unknown process phases, tracking their sequence and duration, and detecting anomalies within these phases. As demonstrated in several industrial applications, the ability of Self-Organizing Maps (SOMs) to detect anomalies, identify and visualize process phases is highly beneficial for comprehensive monitoring and understanding of technical processes. This paper presents a hybrid unsupervised learning strategy (HULS) for monitoring complex industrial processes. Addressing the limitations of SOMs, especially in scenarios with unbalanced datasets and highly correlated process variables, HULS combines existing unsupervised learning techniques to address these challenges. The capabilities of the HULS concept compared to the standard SOM model for the detection of unknown process phases and monitoring process phase sequences and durations are evaluated using a exemplary laboratory batch process.