Abstract:

Signal processing and feature extraction are two of the most significant steps for engine fault di- agnosis. In order to overcome the limitations of the traditional wavelet and morphological wavelet, a new lifting scheme named adaptive morphological gradient lifting wavelet (AMGLW) is presented, which can select between two filters, the average filter and morphological gradient filter, to update the approximation signal based on the local gradient of the analyzed signal. Thus the impulsive components can be enhanced and the noise can be depressed simultaneously by the presented AMGLW scheme. Furthermore, the im- proved non-negative matrix factorization (INMF) is utilized to calculate the features for engine faults clas- sification. The vibration signals acquired from an engine with five working states are employed to validate the proposed engine signal processing and feature extraction scheme.

Key words: information processing, adaptive morphological gradient lifting wavelet, improved non-nega- tive matrix factorization, engine, fault diagnosis