Self-recovered phase reference and structured phase evidence aggregation for bearing degradation transition detection under nonstationary conditions

Degradation-related changes in rolling bearings can reorganize the phase structure of resonance-band responses before pronounced vibration-energy growth. Extracting this information under nonstationary conditions requires a cyclic coordinate that follows the signal and a scoring rule that consolidates phase changes distributed across bands and descriptors. This paper proposes SRR-SPEA, which combines a self-recovered phase reference with structured phase evidence aggregation for bearing degradation transition detection. Baseline records determine the harmonic reference family, cyclic orders and normalization statistics, after which records are processed chronologically using fixed descriptor directions, resonance bands and aggregation settings. Controlled tests covered three nonlinear mechanism families, four noise levels and matched transition and null sequences. In the mixed amplitude-phase benchmark, self-recovery increased the median phase-score Spearman coefficient from 0.340 for a fixed reference to 0.783. In the equal-RMS benchmark, the calibrated phase-score rule detected 60/60 prescribed transitions, produced false alarms in 3/60 null sequences and yielded a median lead of 19.5 segments relative to the prescribed transition center. On full-resolution NASA IMS2-B1, the continuous SRR-SPEA score achieved a Spearman coefficient of 0.838, compared with 0.812 for RMS and −0.082 to 0.188 for the implemented cyclostationary scalar baselines. A sensitive SRR-SPEA event occurred at record 373 before the corresponding RMS event at record 533, whereas a conservative calibration placed the phase-score alarm at record 619. The later IMS bearings and the UPB B01 case further showed that trend coherence and alarm timing depend on degradation trajectory, monitoring resolution and baseline composition. SRR-SPEA therefore provides an interpretable means of detecting and tracing phase reorganization, with its alarm behavior explicitly characterized by calibration and operating conditions.