Vibration-based degradation physics modeling and reliability-driven maintenance optimization of railway axle box bearings using improved fuzzy support vector data description

Understanding the physical degradation mechanisms of axle box bearings and translating them into quantitative reliability metrics is essential for safe railway vehicle operation. This paper presents a physics-informed preventive maintenance (PM) framework bridging vibration-based degradation characterization and reliability-driven maintenance decision-making for axle box bearings. An improved hippopotamus optimization–fuzzy support vector data description model, previously developed for degradation assessment, is extended to map the bearing degradation state—captured through vibration signal features—onto a continuous operational reliability index Rs(t). This data-driven metric replaces the conventional empirically prescribed threshold, providing a physically grounded PM trigger reflecting actual degradation progression. A hybrid failure rate model based on the Weibull distribution, with parameters estimated via maximum likelihood under proportional hazard covariates derived from vibration energy indicators, is employed to formulate unequal-interval imperfect PM cycles under the derived reliability constraint. The accidental failure risk cost, accounting for stage-dependent degradation severity, is incorporated as an economic decision criterion. A full-life-cycle accelerated degradation test on railway vehicle axle box bearings validates the proposed framework. Degradation indicator analysis reveals four physically distinct operational phases: normal, early fault, severe fault, and complete failure. Under the optimal strategy with a reliability threshold of 0.91 and six PM interventions, the maximum average availability reaches 0.9950 and the failure risk cost rate is minimized at 0.000 43. Comparative analysis under three reliability thresholds confirms that maintenance intervals shorten progressively with increasing service time, consistent with fatigue-driven degradation physics, demonstrating the plausibility and applicability of the proposed approach.