Application of wavelet-fractal-multi ART2 neural network to fault recognition of automobile engine

Based on wavelet packet transform and fractal theory, the features are extracted from the non-stationary vibration signals of automobile engine. A self-organizing principle component analysis is used to reduce the dimension of the feature vectors. Then a novel multi-ART2 neural network achieves the classification and recognition of the engine faults. The results prove that the method is efficient.