A fault diagnosis method of rolling bearings based on masking differential iterative filtering and Euclidean mixed entropy

Abstract The capture of uncertainty from nonlinear and nonstationary sensor data is significant for rolling bearing fault diagnosis, which can greatly reduce the probability of rolling bearing failure throughout the life cycle. To address the problem, a feature extraction method based on masking differential iterative filtering and Euclidean mixed entropy (MDIF-EME) is proposed for rolling bearing fault diagnosis in complex environments. Firstly, a MDIF algorithm is presented to extract the random uncertain features and reduce noise interference, where a pair of masking signals is added into the raw signal. Secondly, a EME algorithm based on Euclidean distance between the multiscale permutation entropy and the multiscale dispersion entropy is proposed to detect the performance uncertainty of mode decomposition by capturing both temporal and dynamical features. Finally, the mixed entropy feature set is input into the Gaussian weighted k-nearest neighbor for multi-fault identification. The superiority of the MDIF-EME approach is the fusion of random uncertainty in signals and performance uncertainty in the mode decomposition, which improves noise robustness and feature extraction capability. The experimental results show that the method achieves the best performance with an average accuracy of 95.17% for bearing composite fault diagnosis and outperforms other related methods.