A Health Indicator-Driven Adaptive Feature Mode Decomposition Method for Intelligent Bearing Fault Diagnosis
Vibration signals in rotating machinery are often complex, with fault-induced pulses masked by noise and coupled under compound fault conditions, which increases diagnostic difficulty. Although Feature Mode Decomposition can analyze non-stationary signals, its performance is limited by empirical parameter settings, especially filter length and mode number.A parameter-adaptive framework named EPFMD is developed to address this issue. It optimizes key parameters using a composite health indicator that combines envelope entropy and pulse factor, enabling accurate characterization of fault features. The Ivy Algorithm is applied for automatic parameter optimization. A fusion evaluation index based on kurtosis and pulse factor is then used to select the most fault-sensitive component, followed by envelope demodulation for feature extraction. Validation on the CWRU dataset and experimental data demonstrates that the proposed method effectively identifies inner race, outer race, and compound faults, showing superior performance compared with existing methods.