Development of a vehicle dynamics estimator using a 15DOF model

Forces developing at tire-road interaction provide extremely important information concerning vehicle performance, both for handling and drivability. Such quantities are often utilized during vehicle benchmarking, for the purpose of analyzing the effect of design changes on the behavior of the car. In particular, experimental results can exhibit how tiny design modifications, resulting in equally small variations of the transient built-up of wheel lateral forces, may yield perceptible effect on the overall performance of the vehicle. Unfortunately, the equipment required for measuring such dynamics (wheel force transducers - WFTs) it is very expensive and testing the vehicle for several design variants is a time-consuming process. In addition, current tire simulation models cannot always reproduce such behavior and they usually rely on costly parameter identification processes. In the light of aforementioned issues, the scope of this work of thesis is the development of a vehicle dynamics estimator, which is built over a high-fidelity 15DoF Simcenter AMESim model and which makes use of both IMU and local measurements for estimating wheel lateral loads. Besides this, the thesis aims at investigating whether the usage of such a complex model may be beneficial, as compared to more simple vehicle model (bicycle model and 3DoF model). Given the non-linearity of the observation system, an Unscented Kalman Filter (UKF) is employed for the estimation process thanks to its ability to capture the posterior mean and covariance accurately up to the 3rd order for any non-linearity with no explicit calculation of Jacobians. In particular, this method uses some real measurements (lateral velocity, lateral acceleration, yaw rate, roll rate and wheel rotary velocities) to correct AMESim simulation values in order to estimate wheel cornering stiffness and lateral tire forces. Furthermore, in the final part of this work, the implementation of a recursive gradient based Extended Kalman Filter (EKF) estimator is initiated, by employing the functionalities of the software for linearizing the model and retrieving the Jacobians required by the algorithm. In the future, this can allow to analyze difference in performance between EKF and UKF and to point out advantages and drawbacks of these two implementations. The results of the thesis are then validated and assessed using real wheel loads measurements from different driving maneuvers.