A Method for Extracting Fault Features of Rotating Machinery: Phase Consistency Fusion Entropy

Abstract To address the limitations of conventional Fusion Entropy (FE) in rotating machinery fault diagnosis, including time-frequency energy dispersion, insufficient noise robustness, and limited capability of single-scale analysis to characterize complex signal dynamics, this study first proposes a novel Phase Consistency Fusion Entropy (PCFE) method. Specifically, the Synchrosqueezing Transform (SST) is employed to improve time-frequency energy concentration and resolution, while phase consistency (PC) information is incorporated to weight the time-frequency energy distribution, thereby capturing both energy and phase-structural information. To extend PCFE to multiscale complexity analysis and improve the stability of entropy estimation, a Refined Composite Multiscale (RCM) strategy is subsequently incorporated to establish the refined composite multiscale phase consistency fusion entropy (RCMPCFE). Bayesian optimization is further employed to optimize the hyperparameters of the Support Vector Machine (SVM) classifier for fault identification. Simulation results demonstrate that RCMPCFE exhibits good discriminative capability and stability. On the two real-world fault datasets, RCMPCFE achieves classification accuracies of 99.67% and 99.58%, respectively, substantially outperforming conventional methods.