Rolling Bearing Fault Diagnosis Based on Improved Multi-scale Dispersion Entropy and Deep Extreme Learning Machine

To address the problems of incomplete extraction of rolling bearing fault features and low recognition accuracy, a fault diagnosis model based on local mean decomposition, improved multi-scale dispersion entropy and whale optimization algorithm to optimize the deep extreme learning machine is proposed. Firstly, the model decomposes the original vibration signal into multiple product function components using local mean decomposition; secondly, the spurious PF signal components are removed using mutual information and kurtosis screening criteria; thirdly, the multi-scale fault features of the signal components are extracted by improved multi-scale dispersion entropy; finally, the obtained fault features are input into the deep extreme learning machine optimized by the whale algorithm for training and testing. After conducting experiments with rolling bearings and comparing with the control group, the experimental analysis shows that the model can effectively extract the fault features and improve the identification.accuracy of rolling bearing fault.