Composite multiscale fuzzy extreme position entropy and its application in rolling bearing fault diagnosis

To improve the physical interpretability and multiscale stability of entropy-based methods in nonlinear vibration analysis, a composite multiscale fuzzy extreme position entropy (CMFEPE) is proposed. The method characterizes fault-induced structural distortion by modeling the positional deviation of local extreme via phase space reconstruction. A Gaussian fuzzy membership function transforms discrete extreme position states into continuous probability distributions, alleviating boundary sensitivity. In addition, a composite coarse-graining strategy with multiple starting offsets enhances feature stability across scales. Experimental results on simulated and measured bearing signals demonstrate that CMFEPE achieves superior inter-class separability and fault identification accuracy compared with conventional entropy methods. The proposed method achieves an average classification accuracy of 99.563% on the SDUST data set and 100% on the Polito data set. These results demonstrate the potential of CMFEPE for reliable condition monitoring and early fault diagnosis of rolling bearings under complex operating conditions.