DYNAMIC JOINT ESTIMATION OF VEHICLE SIDESLIP ANGLE AND ROAD ADHESION COEFFICIENT BASED ON DRBF-EKF ALGORITHM

The vehicle center of mass slip angle and the road adhesion coefficient are the key parameters required to realize the intelligence of the vehicle chassis. The vehicle's center of mass slip angle is crucial for improving vehicle safety and handling, and the tire-road adhesion coefficient determines the peak tire force, which in turn determines the vehicle's dynamic stability boundary. In this paper, a dynamic joint estimation method of vehicle mass side slip angle and road adhesion coefficient based on inertial measurement unit (IMU) and built-in speed/angle sensor (WSS) of in-wheel motor is proposed for four-wheel independent drive electric vehicles. The vehicle dynamics analysis of the four-wheel independent drive electric vehicle is carried out, and the vehicle center of mass side slip angle estimator is obtained by combining the Dugoff tire calculation model; the PCA multivariate analysis method is used to reduce the dimension of the high-dimensional data in machine learning, and the principal component characteristic parameters are extracted to establish the road adhesion coefficient estimator. The double radial basis neural network and the extended Kalman filter (double radial basis function and extended Kalman filter, DRBF-EKF) method with an adaptively adjustable network structure are used, the RBF neural network structure is improved by the K-means algorithm, and the extended Kalman filter is used for noise filtering to improve estimation accuracy. Therefore, dynamic joint estimation of vehicle center of mass slip angle and road adhesion coefficient is realized. Simulation and real-vehicle experiments show that the designed DRBF-EKF dynamic joint estimator has better real-time performance and estimation accuracy than the extended Kalman filter algorithm, which can adapt to the changes in road adhesion characteristics and vehicle speed during vehicle driving. Compared with the DRBF method, the estimation accuracy is significantly improved; and the optimal number of neurons in the hidden layer that can meet the estimation accuracy and real-time requirements at the same time is analyzed.