A Reproducible ASM Scenario-Sweep Protocol for Three-Axle Truck Rollover Prediction Using Wheel-Load Features with Independent Validation and HILS

Heavy-duty truck rollovers on curved road sections remain a critical safety concern because lateral excitation generates large roll moments and rapid vertical load transfer across wheels, potentially leading to wheel lift-off. This study proposes a reproducible scenario-sweep protocol in dSPACE Automotive Simulation Models (ASMs) to generate rollover datasets for a three-axle truck and to develop a rollover classifier using physically interpretable wheel/axle vertical load features. Scenarios were parameterized by the vehicle speed (20–30 km/h), curve radius (10–25 m), and bank angle β (0 to −7.5°), with the sign convention defined in this paper) and gross vehicle mass (7000–23,000 kg), where the lateral excitation is governed primarily by ay≈v2R together with the banking contribution. A total of 180 scenarios were used for training, while an independent interpolated validation set of 90 scenarios was constructed using intermediate parameter levels. In offline validation, the proposed model achieved an accuracy of 95.56% (86/90) with no missed-rollover case (FN = 0; errors otherwise consisted of false positives). To assess real-time deployability, the trained pipeline was implemented and evaluated in a hardware-in-the-loop simulation (HILS) configuration over 32 scenarios, achieving an accuracy of 90.63% (29/32) while maintaining FN = 1. By explicitly linking driving-condition inputs to load transfer observables Fz,i(t) and validating the resulting classifier across offline and real-time environments, the proposed workflow provides a reproducible vehicle dynamics grounded pathway for scenario-based rollover risk classification in simulation-to-HILS studies.