A full-process multi-scale numerical framework for earthquake-induced derailment of high-speed train–track–bridge systems

Earthquake-induced derailment of high-speed train–track–bridge (TTB) systems is dominated by strongly nonlinear wheel–rail contact and rapid separation–impact transitions, yet the full-process evolution from guided running to post-derailment impact demand is still rarely quantified in an event-consistent manner. This study presents a full-process, multi-scale framework combining an explicit LS-DYNA model (CRH2 train, CRTS-III slab track, continuous welded rails, and a six-span simply supported bridge) with automated MATLAB post-processing. Derailment is resolved through a geometry-based, irreversible event definition using the wheel vertical displacement relative to the railhead crown as a wheel-drop measure, Δz (positive downward). A unified threshold (Δz = 158 mm) with persistence identifies three instants: peak wheel lift t0 (minimum Δz), derailment onset t1, and first wheel–slab impact t2. Wheel–rail contact forces are converted into comparable envelopes using a zero-phase Butterworth low-pass filter, with the cutoff selected to retain 85% of cumulative PSD energy. After validation against published field and laboratory evidence, the framework is applied to 36 earthquake–speed scenarios (nine motions; 120–300 km/h). Results show that span-end amplification of lateral deck drift drives rail-end plastic bending across expansion gaps, governing derailment initiation in the examined cases. Although absolute triggering times vary among ground motions, the t0–t1–t2 sequence remains near-linear, providing an interpretable event clock for pacing and sensitivity attribution. Conventional force-based indices (Y/Q and unloading ratio U) typically trigger after t1. After guidance loss, lateral velocity continues to accumulate to the first wall impact, reaching 1.77 m/s, which constrains protective-wall demand and provides transferable impact-onset states for refined analyses.