Dynamic Time-Frequency Analysis for Non-Stationary Signal from Mechanical Measurement of Bearing Vibration

The development of manufacturing engineer for aeroengine demands that the monitoring equipment be able to perform in good status, including vibration signal analysis and fault diagnosis. In order to acquire the decisions in accordance with experiment result, one must have a powerful tool of signal feature extraction for fault pattern recognition, which has significant effect on sampled data processing. The wavelet transformation can satisfy transient signal requirements and is applied in representing sampled data or other functions at different scales or resolutions. The wavelet network is introduced as a class of feedforward networks consisted of wavelets, in which the wavelet transformation is utilized for analysis of neural network. The frequently used method is to construct multidimensional mother wavelet by compositing the single dimensional scaling function and wavelet in different dimensions in the tensor product. The generic algorithm is used to complete the parameter determination of wavelet network, acquiring fast convergence speed. The experiment result demonstrates that the combination of wavelet transformation with neural network can remedy the weakness of each other, resulting in network with efficient construction method and fault pattern recognition in good performance.