Abstract
In high-speed railways (HSR), rail over-grinding degrades the dynamic performance of vehicles, often causing the carbody lateral acceleration (CLA) to exceed safety limits. Rail grinding is an effective mitigation measure, but current engineering practices lack models for formulating precise grinding strategies. To address this gap, this study focuses on the carbody sway section of CRH380B EMUs operating on the Xulan High-Speed Railway. It collects and analyzes over-ground rail profiles, which are fitted using the GA-NURBS curve. Combined with a refined vehicle–track coupled dynamics model, the study innovatively proposes the Gravitational Search Algorithm-Particle Swarm Optimization-Back Propagation-Non-dominated Sorting Genetic Algorithm III-Simulated Annealing (GSA-PSO-BP-NSGA Ⅲ-SA) hybrid optimization algorithm—tailored to meet high-dimensional optimization requirements—yielding an optimized target grinding profile. Additionally, based on full-sample grinding data, multiple typical grinding strategies corresponding to distinct profile characteristics are summarized. The study further innovatively constructs the Intelligent Grinding Strategy Model (IGSM) driven by the Regularized Multi-Output Bidirectional Long Short-Term Memory-Attention-Kernel Extreme Learning Machine (RBMO-BiLSTM-Attention-KELM), realizing accurate mapping from measured profiles to optimized strategies. Results show that the optimized profile reduces CLA by 14.3%, bogie frame lateral acceleration by 12.5%, derailment coefficient by 33.3%, and fatigue index by 52.6%. This effectively controls the grinding depth and improves wheel–rail contact. The IGSM enables the real-time output of optimized strategies, making it suitable for practical engineering applications. After field application, the section showed no CLA exceedances or rail surface fatigue damage within 3 months, with rail profiles consistently meeting acceptance standards.
Keywords
Get full access to this article
View all access options for this article.