Simulator-recovery assessment of a physics-informed hybrid framework for tire wear rate estimation from telematics data

Route-level tire wear estimation from telematics data could support non-exhaust-emission assessment and sustainability-oriented navigation, but navigation-derived features do not directly observe the tire–road contact variables that control material removal. This paper presents a methodological simulator-recovery framework—not a validated predictive model—that links navigation-app features to an Archard-type tire wear simulator. The framework’s main contribution is a transparent input taxonomy that separates deployable quantities (available at inference) from oracle quantities (available only to the synthetic label generator), together with a dimensionally consistent Archard formulation, a per-tire normal-force model with longitudinal, lateral, and grade load transfer, a saturated slip-ratio model with low-speed correction, and a differentiable physics layer trained with an algebraic Archard ratio-consistency loss. On 150 synthetic routes (3588 telematics segments), pure kinematic data-driven models perform poorly (Random Forest: R 2 = 0 . 09 ; Gradient Boosting: R 2 = 0 . 00 ) because route-specific wear-coefficient variation is intentionally unobserved. Upper-bound reference diagnostics that use label-generation physics information reach much higher apparent accuracy: Hybrid Physics+RF reaches R 2 = 0 . 97 and MAE = 4.47 mg/km, while the physics-only reference reaches R 2 = 0 . 98 . These values should be interpreted strictly as simulator-recovery upper bounds, not as evidence of real-world predictive capability or deployable tire-wear prediction. A sensitivity analysis identifies the wear coefficient K as the dominant source of simulator variability (a consistent 490% range across the manuscript), highlighting tire-parameter calibration as the central deployment barrier. A loss-weight sensitivity analysis, a cross-vehicle transferability diagnostic, and an explicit discussion of deployable versus oracle quantities are provided to clarify the scope and limits of the framework. The study is positioned as a methodological starting point for future machine-learning work with measured telematics and tire-wear data.