Estimation of truck mass and center of gravity using data-driven mechanics: a comparative study of TSO-SVM, CS-BP, SSA-ELM, and WOA-XGBoost in automotive dynamics

Accurate estimation of a truck’s mass and center of gravity (CG) is critical for optimizing safety and performance but remains challenging due to dynamic uncertainties in weight distribution and road interactions. This study introduces a data-driven mechanics framework integrating four hybrid machine learning (ML) models-tuna search-optimized support vector machine, cuckoo search-optimized BP neural networks, sparrow search algorithm-optimized extreme learning machine, and whale search-optimized XGBoost-to enable estimation. A 17-degree-of-freedom multibody dynamics model, incorporating suspension kinematics via a semirecursive formulation, generates simulation datasets linking real-time tuck states (pitch, roll) to mass and CG. Search algorithms leverage physics-derived truck state data to initialize ML hyperparameters, enhancing training efficiency. Validation against multibody benchmarks confirms accuracy, while robustness is demonstrated across driving scenarios and noise. By unifying data-driven ML with physics-based mechanics, this approach advances parameter estimation, bridging truck dynamics with computational intelligence for automotive design.