Real-Time Mass Estimation of Commercial Trucks Using a Wheel-Dynamics-Based Disturbance Observer with Data-Driven Parameter Optimization

Accurate vehicle mass information is essential for improving the safety, energy efficiency, and motion-control performance of commercial trucks. In practical applications, however, direct mass measurement is rarely available during operation, and payload variations can significantly degrade acceleration prediction, braking allocation, traction control, and advanced driver-assistance functions. This paper proposes a real-time mass estimation algorithm for commercial trucks using a wheel-dynamics-based disturbance observer with data-driven parameter optimization. The proposed method exploits wheel rotational dynamics to reconstruct longitudinal force-related quantities from wheel-speed measurements and wheel-torque-related information, and estimates the vehicle mass through a real-time filtered force-to-acceleration relationship using the measured longitudinal acceleration. The data-driven component is formulated as an offline calibration stage that identifies a fixed parameter vector from real-vehicle Controller Area Network (CAN) data; it does not replace the physical wheel model or perform online parameter learning. A signal-conditioning and supervisory implementation layer, including input sanitization, wheel-torque filtering, braking-condition reset, validity gating, resistance compensation, and moving-average smoothing, is used to prevent unreliable mass updates under invalid or weak-excitation conditions. The algorithm is implemented on an Infineon AURIX TC375 automotive microcontroller and validated through repeated real-vehicle commercial-truck experiments under six payload conditions at 50 km/h and 80 km/h. The average final estimation errors were 16.97% and 9.55% at 50 km/h and 80 km/h, respectively. The corresponding 43.8% reduction refers only to the difference between the two experimental conditions considered in this study and is not a comparison with previously published methods. The experiments show repeatable payload-dependent estimation behavior, while also revealing a systematic underestimation tendency under heavily loaded conditions.