Selection of wavelet packet basis for rotating machinery fault diagnosis

Wavelet packets provide a library of orthonormal bases and can therefore represent a signal in many different ways. This paper presents a method to select the orthonormal basis to represent vibration signals for rotating machinery fault diagnosis. The selected basis is formed by two sets of basis functions for representation of the transients excited by localized faults and other signal components of interest, respectively. This method overcomes the limitation of the widely used Coifman and Wickerhauser's best basis algorithm in detection of transients hidden in large background vibration and is more effective in rotating machinery fault diagnosis especially in the early stage of fault development. In addition, differing from some existing studies, this method does not need training samples and is relatively easy to implement in practice. Applications in the analysis of gearbox and bearing vibration signals show that the proposed method performs better than the best basis algorithm.