Rolling bearing fault diagnosis based on three-domain feature extraction and WOA-ELM

In order to effectively raise the fault detection accuracy of rolling bearing, a method is proposed in extracting different domains′ feature and optimizing extreme learning machine (ELM) by whale algorithm. Firstly, time domain analysis, spectrum analysis and wavelet packet decomposition are used in the process of the vibration signal. Secondly, in order to avoid dimensional disasters, locality preserving projection in manifold learning is applied to reduce the dimensions of mixed feature sets and eliminate redundant features; In order to ameliorate the deficiency that ELM is prone to come to local optimal value, whale optimization algorithm is used to adjust the parameters of network; Lastly, WOA-ELM rolling bearing diagnosis model is established to classify and diagnose faults. Using the bearing data from Western Reserve University to simulate. The result of the test shows that this method can usefully increase rate accuracy of fault diagnosis.