Spectral-Trend Adaptive Amplitude-Period Weighted Ordinal Pattern Mode Decomposition for Bearing Outer-Race Fault Feature Enhancement

Bearing outer-race fault impacts under strong noise and periodic interference are difficult to separate because ordinal-pattern anchors can be dominated by non-fault impacts and because a fixed embedding scale cannot match resonance bands of different widths. This paper proposes spectral-trend adaptive amplitude-period weighted ordinal pattern mode decomposition (ST-APWOPMD) for outer-race fault feature enhancement. A weighted spectral trend is extracted from the normalized logarithmic spectrum, and its valleys partition the candidate bands; band energy, waveform kurtosis, and envelope kurtosis are combined into a blind score for band selection. The PRSA half-window length and the ordinal embedding delay are inferred from the bandwidth and center frequency of the selected band. A joint weight that combines the local envelope amplitude with the recurrence at one impact period is applied to the ordinal-pattern statistics and to the PRSA template. A band-constrained conservative update then refines each mode. The method is compared with OPMD, VMD, and EEMD on a simulated outer-race signal, a segment of the SJTU-SY test-bench data, and a segment of an urban-rail wheelset axlebox test rig. For these three examples, the fault feature index (FI) of ST-APWOPMD is 4.704, 2.347, and 2.625, respectively, the highest among the four methods; the corresponding raw-signal FI values are 3.250, 2.069, and 0.779. The single-run times are of the same order as OPMD and much shorter than the EEMD runs reported here. The validation is limited to constant-speed outer-race records; fault classification, other fault types, variable-speed operation, and field data are not addressed. A 20-segment axlebox study and a component-wise ablation identify band selection as the dominant component.