Bayesian neural network-driven accelerometer-based type intelligent tire force measurement system

Accurate real-time measurement of tire forces is crucial for vehicle dynamics control. However, the current cost of direct tire force measurement is very high. In this study, we propose a cost-effective tire force measurement method based on in-tire acceleration measuring information and Bayesian Neural Network (BNN). To develop this measurement system, we designed three key components: (1) Fourier Amplitude Sensitivity Test (FAST) was used to determine the optimal accelerometer arrangement for improving this system measurement accuracy; (2) Signal preprocessing algorithms was designed to extract promising acceleration signal features of tire force; (3) Bayesian Neural Network was used to achieve precise tire force estimation. Experimental results indicate that the estimated tire force of this system has a good agreement with the reference tire force under varying conditions, including load, slip ratio, slip angle, tire pressure, vehicle speed and road friction coefficient. The offline Mean Absolute Percentage Errors for tire longitudinal, lateral, and vertical forces based on the BNN were 2.64%, 2.44%, and 0.55%, respectively. The online estimation also demonstrated good validation results. Additionally, under high-friction conditions, the BNN tire force estimation produced smaller confidence intervals, indicating greater stability.