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

In response to the challenges associated with the incomplete extraction of fault features in rolling bearings and the resultant low recognition accuracy, this study proposes a fault diagnosis model that integrates local mean decomposition, enhanced multi-scale dispersion entropy, and a whale optimization algorithm to refine the deep extreme learning machine. Initially, the model employs local mean decomposition to break down the original vibration signal into several product function components. Subsequently, spurious product function signal components are eliminated through the application of mutual information and kurtosis screening criteria. The multi-scale fault features of the signal components are then extracted using the improved multi-scale dispersion entropy method. Finally, the extracted fault features are fed into a deep extreme learning machine, which has been optimized via the whale optimization algorithm, for training and testing purposes. Experimental investigations conducted on rolling bearings, alongside comparisons with a control group, demonstrate that the proposed model effectively extracts fault features and enhances the identification accuracy of rolling bearing faults.