The Development of A Novel Fault Identification Technique by Combining Minimum-Distance Pattern-Recognition and Discrete Wavelet Transform

In this paper, a new idea in machinery fault diagnostic by combining minimum-distance pattern-recognition and discrete wavelet transform (DWT) is presented. The method starts by forming the feature vector, which is basically the position vector of a point in the hyperspace spanned by the set of wavelet coefficients. These coefficients are obtained by expanding the measured signal from faulty machines using DWT. Fault identification is performed based on the distance between the feature vector of the measured signal and the reference feature vector. The basic idea comes from the fact that the wavelet expansion coefficients are unique for a particular signal, considering completeness of the wavelet series. In order to show the effectiveness of the method, two standard machinery faults, namely the unbalance case and the mechanical looseness, were simulated. Experimental results are very promising, since the method successfully identifies the simulated faults.