A Comparison of Model-Based and Machine Learning Techniques for Fault Diagnosis

Abstract

In recent years, there has been a lot of interest in Fault Detection and Isolation (FDI) for systems. Model-based methods and Machine Learning (ML)-based approaches have been extensively developed to detect and identify specific faults by taking into consideration, respectively, the mathematical description of the monitored process and the statistical model constructed from historical data. Recently, studies have been conducted to combine both approaches to improve FDI performance. This study provides a side-by-side comparison of both approaches on the same system, which will aid in determining the best way to combine both approaches to create a hybrid FDI. First, the current state of the art in model-based, ML-based, and hybrid FDI is reviewed. Second, the detailed experimental setup and principles of both FDI approaches are discussed. The FDI of an actual Storage Device (SD) utilized in a green hydrogen production platform is then performed using both methodologies. Finally, it is stated that while both approaches have advantages and disadvantages, they can be combined to complement each other and improve the FDI performance.