To address the complex and highly variable wheel–rail adhesion behavior on high-speed railway curves, this study establishes a numerical wheel–rail rolling contact model based on starved elastohydrodynamic lubrication (EHL) theory and Herschel–Bulkley rheological characteristics. The model validation yielded RMSE = 0.0228, MAE = 0.0217, MAPE = 11.80%, R² = 0.828, and a 95% confidence interval of the mean residual of −0.0298 to −0.0136. The study focuses on the initial operational phase after application, systematically quantifying the fluid-dynamic regulation mechanisms of water-based friction modifiers once a thin, starved lubricating film has been formed on the rail surface under curving conditions. By analyzing rail profiles (CHN60 and CHN60N), operating parameters, and track geometry, this study shows how adhesion behavior on curved track sections is governed by the coupled effects of contact mechanics and lubrication. As the outer rail superelevation increases from 0 to 70 mm, the adhesion coefficient decreases by approximately 15–25%, mainly because the reduced normal force shifts the wheel–rail interface toward the Stribeck transition regime. Increasing axle load from 14 t to 30 t reduces the dimensionless film thickness, but the enlarged contact area contributes to a more stable adhesion level, with an increase of about 12%. Compared with the CHN60 profile, the CHN60N profile exhibits better geometric conformity, producing a lubricating film that is 10–15% thicker and leading to a lower and more stable adhesion coefficient, decreasing from approximately 0.35 to 0.1. The results also identify a critical lateral displacement of around −4 mm, beyond which the contact radius becomes stable and the adhesion coefficient reaches a minimum plateau. These findings clarify the competing effects of fluid entrainment and metallic asperity contact, and provide quantitative guidance for friction management and friction modifier application on curved track sections.

Top-of-rail friction modifiers (TORFMs) act as transient interfacial films that regulate wheel–rail adhesion under rolling–sliding contact. Unlike permanent coatings, these films undergo continuous formation, depletion, and regeneration during repeated wheel passages. This study investigates TORFM-treated interfaces by combining laboratory rolling-contact tests with a starved non-Newtonian elastohydrodynamic lubrication (EHL) model. Experimentally, application dosage and viscosity were varied on a CL60 wheel–U71Mn rail pair, and the adhesion coefficient was measured as the validation quantity. Numerically, a model-based sensitivity analysis was used to examine how dosage-derived effective supply thickness, viscosity, and rheological index influence film formation and load sharing. The experiments show that increasing dosage alleviates inlet starvation and lowers initial adhesion, which then recovers toward dry-contact levels during repeated rolling. Increasing viscosity from 0.002 to 0.022 Pa·s reduced the minimum adhesion coefficient from 0.21 to 0.086, although excessive viscosity may prolong the low-adhesion stage. The rheological-index calculations (n = 0.7–1.0 around the measured formulation value n = 0.82) indicate that shear-responsive flowability can affect contact-zone supply and the film-supported load fraction; these rheological-index trends are model-based sensitivities rather than independently validated experimental effects. Overall, the results support a transient interfacial-film interpretation of TORFM-controlled adhesion and suggest that practical design should target a controllable medium-adhesion window (approximately 0.15–0.30 under the present laboratory conditions) rather than minimum friction alone.