Abstract
In accordance with the characteristics of key parts of automobile transmission system, this paper discusses the whole process of fault diagnosis based on vibration signals, including feature extraction, pattern recognition and diagnosis decision. To stable signals, feature extraction is fulfilled by various methods in time domain and frequency domain. To time varying signals, feature extraction is fulfilled by wavelet transform. Feature vectors are composed subsequently. Neural networks and grey system theory are used to diagnose fault in single sensor mode separately. On these bases, in order to improve the precision of decision, data fusion technique is used to deal with the information from multi-sensors comprehensively.