Wavelet Novelty Measure for Machinery Diagnostics

The paper is concerned with wavelet analysis for fault detection in machinery diagnostics. A new approach based on novelty detection is presented. The method involves a wavelet compression algorithm to vibration data in order to extract a set of features which are related to the fault. The compression algorithm uses orthogonal Daubechies’ wavelets and a simple thresholding procedure. The wavelet based novelty measure is established as a statistical distance between decoded data representing different fault advancements.