Bearing Fault Diagnosis Using Modified Multi-scale Sample Entropy and One-against-rest Feature Selection

For bearing fault diagnosis, how to find and select sensitive features that can improve the diagnostic accuracy is a key problem. To this end, a new fault diagnosis approach based on modified multi-scale sample entropy (MMSE) and one-against-rest feature selection is proposed to identify different bearing faults. Firstly, several MMSEs as improved fault features are constructed to find informative fault information from the raw signals. Subsequently, a new feature selection method based on the one-against-rest strategy is proposed and used to select the fault features. Finally, the selected important MMSEs are fed into a naive Bayesian classifier to perform fault identification. Experimental results demonstrate that the proposed method can effectively diagnose bearing faults with high diagnosis accuracy.