Distributed Drive Electric Vehicle Longitudinal Velocity Estimation with Adaptive Kalman Filter: Theory and Experiment

Velocity is one of the most important inputs of active safety system such as ABS, TCS, ESP et al. Sensors which can measure velocity directly are still expensive and not widely assembled in production automobiles. In a distributed drive electric vehicle equipped with four in-wheel motors, velocity is hard to obtain due to all wheel drive, especially in wheel slip conditions. This paper focus on longitudinal velocity estimation of the distributed drive electric vehicle. Firstly, a basic longitudinal velocity estimation model is built based on a typical used Kalman filter, where four wheel speeds obtained by wheel speed sensors constitute an observation variable and the longitudinal acceleration measured by an inertia moment unit is chosen as input variable. In simulations, the typical Kalman filter show good results when no wheel slips severely; when one or more wheels slip, the typical Kalman filter with constant covariance matrices does not work well, while estimation results can be improved by calibrating observation error covariance matrix and/ or system error covariance matrix. Therefore, the adjusting gain matrix-based Kalman filter which can detect the wheel slip and cope with that is proposed. Simulation in different conditions are carried out, including no wheel slip, one wheel slips, all wheel slip, passing bump, as well as long time acceleration and deceleration on rough road, and the results show that wheel slip have very little impact on estimation velocity. On-road experiments, including drive with sudden acceleration and deceleration, past a bump, and accelerate on wet tile road, show satisfying results. On wet tile road, where the maximum slip rate is larger than 0.9, the velocity estimation error converges to within 5% in 1s and to zero at last.