Diagnosis of PEMFC by using statistical analysis

International audienceFault diagnosis, especially on-line fault diagnosis is an essential issue for practical application of Polymer Electrolyte Membrane Fuel Cell (PEMFC) system. This paper proposes a diagnosis approach for PEMFC to handle the flooding fault which is considered to be acommon fault. In this procedure, both fault detection and fault isolation are considered. For fault detection, the statistical characters of cell voltages distribution of a 20-cell PEMFC stack are analyzed. Parameters for describing voltage distribution characters are extracted.After that, a subset of parameters is selected in the orientation that the definition of fault is as correct as possible. A popular clustering methodology named K-means clustering (KMC) is adopted to make definition of flooding fault zone. For fault isolation, Support vector machine (SVM) classifier is trained to handle the cell voltages constructed vectors. Two different causes of flooding: increasing air humidity and decreasing of stack temperature can be discriminated by the classifier with a high correctness