Vehicle Dynamics State Estimation Based on Sensor Fusion by Adaptive Kalman FilterDoctoral thesis, Faculty of Electrical Engineering and Computing, University of Zagreb, Croatia, 2015
An increasing number of vehicle dynamics control systems are being embedded into modern vehicles in order to assure safety and comfort of driving. All of these systems require information on the vehicle dynamics state variables (e.g. yaw rate, sideslip angle, roll angle etc.). Some of them can be measured, while others need to be estimated based on available measurements and appropriate vehicle kinematics/dynamics models. This thesis presents a contribution to the research of yaw rate and sideslip angle estimation. More specifically, a kinematic sensor fusion-based yaw rate estimator has been proposed, which combines the wheel speeds measured by standard Anti-lock Braking System (ABS) sensors and the measurement of vehicle lateral acceleration obtained from two accelerometers placed diagonally upon the chassis. Similar fusion concept has been employed for development of a kinematic vehicle sideslip angle estimator utilizing information obtained by low-cost inertial sensors and single-antenna GPS receiver. Moreover, a sideslip angle estimator based on vehicle dynamics model with stochastic modeling of the tire forces has been proposed and used for concurrent estimation of other vehicle dynamics variables and parameters, such as the tire sideslip angles, lateral tire forces, tire cornering stiffness, and tire-road coefficient of friction. The research methodology includes: setup of appropriate kinematic and/or dynamic vehicle models; identification, open-loop compensation, and analysis of dominant sources of estimation errors; and design of estimators based on the sensor fusion principle by using the adaptive extended Kalman filter. Verification of the developed estimators has first been carried out by means of computer simulations based on an experimentally verified ten degrees-of-freedom vehicle dynamics model comprising the magic-formula tire model. In the case of dynamic sideslip angle estimator with stochastic tire modeling, the estimation accuracy has also been verified experimentally, based on the data recorded on a test vehicle equipped with a high-precision inertial measurement unit and two-antenna GPS receiver, as well as by using a standard set of vehicle dynamics control system sensors. In order to obtain a favorable performance of the vehicle state variable estimation under the various operating conditions, a rule-based adaptation of the Kalman filter state covariance matrix has been utilized for kinematic estimators, while for the dynamic, model-based vehicle sideslip angle estimator an adaptive fading algorithm has been implemented for adaptation of the Kalman filter state and measurement covariance matrices.