Analysis of vibration signals for condition monitoring of power switching equipment using wavelet transform

A transient vibration signal processing technique jointly utilising wavelet transforms and a neural network has been developed. A nondecimated wavelet transform is employed as the wavelet band energy feature extractor and decimated wavelet transform as the envelope feature extractor for primary vibration signals. The feature patterns prepared from the output of the two feature extractors are used to train a self-organising map (SOM) for fault detection on power switching equipment. From the present investigation, it is concluded that wavelet-based feature extractor permits the extraction of essential features from transient vibration signals. Normalised quantisation error (NQE) of SOM is introduced to quantify the deviation of a vibration signal and its template. Numerical results show that NQE can be used as an effective indicator for faults occurring in a power switching device.