A Hybrid Unsupervised Learning Strategy for Monitoring Complex Industrial Manufacturing Processes

Industrial production processes are complex systems that require continuous monitoring to ensure efficiency, product quality, and safety. Anomaly detection is a crucial component of these monitoring systems, with machine learning (ML) methods offering significant advantages over traditional statistical techniques. Clustering-based unsupervised anomaly detection algorithms, such as Self-Organizing Maps (SOMS), are particularly valuable in process monitoring, as they can detect anomalies and identify process phases without requiring labeled training data. However, unsupervised learning methods can be sensitive to the inherent properties of the training data, particularly in cases of strong correlation between features or unbalanced datasets. This paper presents a hybrid unsupervised learning strategy (HULS) for monitoring complex industrial production processes. The proposed method combines existing techniques to address the challenges posed by correlated and unbalanced data. Experimental results from a synthetic dataset demonstrate the effectiveness of the developed hybrid approach.