LAUSR

In this paper, data-driven approaches are applied to identify faults of a practical polymer electrolyte membrane (PEM) fuel cell system. Signal processing approaches are selected and employed to multiple sensor measurements, including methodologies reducing the dimension of the original dataset, and techniques extracting features. Both supervised and unsupervised techniques are applied in this study to investigate the robustness of the diagnostic procedure. Moreover, due to the fact that a series of features can be extracted from these sensors, the singular value decomposition (SVD) technique is applied to select features providing better diagnostic performance. Results demonstrate that with features selected from SVD, fuel cell system faults can be detected more effectively, and various fuel cell faults can also be discriminated with good quality. From the findings, conclusions are made and further work is suggested.