Two-stage least square method for model identification of vehicle motion

Abstract

Vehicle dynamics have very complex characteristic and nonlinear behaviour. Vehicle dynamics are decomposed of many internal and external components which influence vehicle stability. External components come from environment such as wind forces, surface coarse of road, lane bend or sudden maneuver, which will change the value of vehicle stability parameters, i.e. yaw rate and sideslip. Both are influenced by the longitudinal velocity change and are difficult to be measured by installed sensors in vehicle. For driving convenience and high safety performance, the vehicle stability parameters must be controlled. Researches and experiments directly on the vehicle bring quite expensive cost and huge time consuming. Therefore, before doing experiments to the real vehicle, simulation is taken. Simulation needs model of vehicle dynamics that are approaching real vehicle dynamics. In this paper, instead of using simple vehicle model, the replication of the vehicle dynamics has been taken from CarSim multi-degree of freedom vehicle model. CarSim's vehicle model C Class Hatchback Sprungmass 2012 is used in this simulation. All vehicle parameters are already provided by CarSim. Vehicle model run along defined part of vehicle track of Universitas Indonesia. At certain bend lane, the obtained data consists of steering angle, longitudinal forces to all four wheels, yaw rate and side slip angle. Two-stage Least Square method has been applied to those data in order to estimate vehicle dynamics. The estimated model was validated upon another data. The result shows that the estimated vehicle model could represent in approaching real vehicle dynamics. The estimated model has perfect controllable and observable characteristic. The model is stable and its eigenvalues is inside unit circle.