An effort on the fault diagnosis for the final drive assembly with the characteristics in course and spectrum

It has been acting as the standard process on evaluation of the final drive assembly in automotive that the operator gives the results from noise based on their experience. Obviously the clues about to faults also depend on the vibration of the final drive. There are great advantages to get the noise vibration signal in the fitting shop. A method for fault diagnosis of the quality of automotive final drive assembly based on wavelet- neural network with mixing characteristics of vibration signal is presented in this paper. The vibration signals of final drive acquisition system are preprocessed to extract the properties in time domain and wavelet transform is used to decompose the signal into eight vectors in different frequency bands. The energy vectors of eight features extracted from the wavelet transform and other from course are used as inputs to the artificial neural network (ANN) in the diagnosis system. The ANN is trained according to back-propagation (BP) algorithm with a subset of the experimental data from known assembly conditions. The ANN is tested with the other set of unknown assembly conditions data. The results obtained indicate the effectiveness of the extracted features from course and spectrum and the effective classification of ANN in diagnosis of the quality of final drive assembly.