• Proposes a novel rolling bearing early fault detection method using a feature clustering fusion degradation index, which integrates multidomain statistical features to improve detection accuracy.
• Introduces an improved hierarchical clustering algorithm combined with feature evaluation indices to select a preferred feature subset, reducing redundancy while preserving information richness.
• Designs an improved exponential weighted moving average (EWMA) control chart with accurate parameter evaluation and anomaly determination strategy, enhancing sensitivity to small fluctuations from early faults.
• Validates the method on public datasets, demonstrating its effectiveness in providing comprehensive degradation indicators and eliminating interference from random anomalies and running-in periods.