Performance of wavelet analysis and neural network for detection and diagnosis of rotating machine fault

A novel approach combining wavelet transform with neural network is proposed for vibration fault diagnosis of turbogenerator set in power system. The multi-resolution analysis technology is used to acquire the feature vectors which are applied to train and test the neural network. Feature extraction involves preliminary processing of measurements to obtain suitable parameters which reveal weather an interesting pattern is emerging. The feature extraction technique is needed for preliminary processing of recorded time-series vibrations over a long period of time to obtain suitable parameters. The neural network parameters are determined by means of the recursive orthogonal least squares algorithm. In network training procedure, much simulation and practical samples are utilized to verify and test the network performance. And according to the output result, the fault pattern can be recognized. The actual applications show that the method is effective for detection and diagnosis of rotating machine fault and the experiment result is correct.