Purpose <p>Early fault signals in high-speed train axle box bearings are easily buried in strong noise, which leads to a barrier for high detection efficiency. Based on the Co-Frequency Convex Peak Method (CFCPM), the study aims to develop a faster approach for weak signal detection.</p> Methods <p>By introducing asymmetric damping into a nonlinear system, an Asymmetric Damping Scale-varying Detection System (ADSDS) is constructed. The range of directional transition thresholds of ADSDS in the periodic state is theoretically determined using the harmonic balance method and the Melnikov method. The relationship between the potential-well transition rate of ADSDS and the noise intensity is theoretically calculated via the Fokker–Planck Equation (FPE). Depending on the binomial fitting of numerical simulation data, an analytical estimation expression for determining the transition threshold of ADSDS under noisy conditions is obtained. Furthermore, leveraging the directional transition characteristic of ADSDS in the periodic state, a Periodic State Directional Transition Method (PSDTM) is developed for rapid weak signal detection.</p> Results <p>Numerical simulations confirm that ADSDS is sensitive to target-frequency signals and suppresses non-target frequencies. Experimentally, PSDTM is validated using vibration data from artificially faulted axle box bearings collected via a roller rig, which reduces bifurcation diagram generation time by 160 times compared to CFCPM.</p> Conclusion <p>PSDTM significantly improves detection efficiency while maintaining accuracy, which promotes the development of early fault diagnosis in high-speed train axle box bearings.</p>