Hybrid intelligent fault diagnosis based on adaptive lifting wavelet and multi-class support vector machine

To diagnose compound faults of rotating machine, this paper presents a novel hybrid intelligent fault diagnosis model based on adaptive lifting wavelet and multi-class support vector machine. First of all, the adaptive lifting wavelet is constructed to mach the signal local characteristics. The original signal is decomposed into approximation signal and detail signal. Secondly, 32 time-domain statistical features are evaluated and some salient features are selected from them by applying the distance evaluation technique. Finally, multi-class support vector machine (SVM) is applied. The testing classification accuracy with salient features of the proposed model reaches to 98.32%, which is 5.32% and 5.04% higher than classification with the salient features of original signal and the salient features of approximation signal and detail signal decomposed by second generation wavelet. It shows that the proposed model can effectively lock signal local characteristics, recognize different fault categories and enhance classification accuracy.