A review on non-model based diagnosis methodologies for PEM fuel cell stacks and systems

A review of non-model based methodologies applied to diagnosis of Proton Exchange Membrane Fuel Cell (PEMFC) system is presented. Three types of non-model based methods including artificial intelligence, statistical method and signal processing method are discussed and compared. The artificial intelligence one, divided into Neural Network (NN), Fuzzy Logic (FL) and neural-fuzzy method, is applied as a fault classifier which is quite different from its role in model-based method. Linear feature reduction methods including Principle Component Analysis (PCA) and Fisher Discriminant Analysis (FDA), and nonlinear ones such as Kernel PCA (KPCA) and Kernel FDA (KFDA) are demonstrated as part of statistical methods. Additionally, a statistical theory based classifier- Bayesian Network (BN) is also introduced in this part. As for signal processing method, both Fast Fourier Transform (FFT) for stationary signals and short-time Fourier Transform (STFT), as well as Wavelet Transform (WT) for non-stationary signals are introduced. Since each method has its advantages and limitations, a comparison is made finally and hybrid approaches resulting from integration of different methods are believed to be promising.