Learning-based approaches for automatic fault detection and diagnosis in industrial systems

The timely detection and diagnosis of faults are crucial for maintaining the safety and reliability of industrial systems, as they help prevent severe damage and unexpected disruptions in operations. In pursuit of this objective, learning-based approaches have emerged as powerful tools, harnessing various machine learning techniques to detect potential faults and diagnose their root causes in the systems under study. This thesis delves into the application and advancement of machine learning methods, especially in the context of hydraulic and pneumatic systems, vital components at the heart of industrial machinery. The primary focus is to enhance the fault detection and diagnosis capabilities within these critical domains, thereby contributing to the overall performance and longevity of industrial machinery. The hydraulic system serves as a primary focus of investigation, where an early time series classification method is applied to detect faults as early as possible. Leveraging data-driven approaches, the aim is to detect potential faults in the multi-component hydraulic system and diagnose their underlying causes using a multi-class multi-output classification method. The study encompasses the development of algorithms capable of recognizing deviations from normal system behavior and, in turn, determining the specific issues that trigger these deviations. Moving forward, my exploration extends to both pneumatic and hydraulic systems in case of label scarcity in the dataset. To this aim, semi-supervised learning approaches are combined with conventional classification methods to harness the power of unlabeled data and improve model generalization and performance in fault detection and diagnosis. Finally, by employing adaptive machine learning methods in this context, an adaptive constraint clustering algorithm is presented for real-time fault detection in the pneumatic system. The results of this thesis are anticipated to provide practical solutions for maintaining the safety and reliability of complex industrial systems.