Supervised classification approach dedicated to proton exchange membrane fuel cell diagnostic

This paper deals with a supervised classification approach dedicated to diagnose a PEMFC system. The purpose is to detect and isolate a fault occurring on a fuel cell system based on electrochemical impedance measurements. The air compressor failure, fuel poisoning and stack overheating are the faults considered in this study. Some experimental tests are performed on two similar stacks (only the power changes). The k-nearest neighbors is the supervised classification method that is used as it is well known for its efficiency and simplicity of implementation.