A wavelet transform-artificial neural network (WT-ANN) based rotating machinery fault diagnostics methodology

This paper outlines a Wavelet Transform (WT) based Artificial Neural Network (ANN) input data pre-processing scheme and presents the results of localized gear tooth defect recognition tests by employing this proposed methodology. The methodology consists of calculating Daubechies’ 20-order (DAUB-20) mean-square dilation WTs of the data, and then selecting predominant wavelet coefficients distributed to certain levels of these WTs as inputs to ANNs for pattern recognition. The test results show that a fairly small sized backpropagation network trained with a reasonably small number of training sets can detect and classify various types or degrees of failures occurring on a spur gear pair successfully.