An intelligent control of vehicle dynamic systems by artificial neural network

In this study, two artificial neural network controls for a vehicle four wheel steering system, a sliding mode control combined with an artificial neural network and an unsupervised learning control, have been proposed. In the first control scheme, the neural network estimates known or even unknown dynamics such that the control parameters of the sliding mode can be adaptively adjusted. The adaptive capability of neural network minimizes the necessary switching gain of the discontinuous control to compensate for the uncertainties. This control scheme generates less chattering than the conventional sliding mode control, resulting in reduced steady state errors as determined by an approximation technique. Lyapunov's direct method is employed to guarantee both the stability and the level of performance of the global system.; The second control scheme, a neural controller using unsupervised learning, is able to compensate for the unknown uncertainties, resulting in the robust control of nonlinear vehicle dynamics. The teaching signal for the training is the difference between the actual plant output and the reference model output. This control scheme does not require a knowledge of the inverse dynamics of the plant or the Jacobian information of the learned plant. As a result an on-line control can be carried out.; In order to describe the dynamics of a 4WS (Four Wheel Steering) vehicle, a three degree-of-freedom (3 DOF) vehicle handling model is used. A neural network tire model is used for identification of the highly nonlinear tire side force, a dominant external force to the vehicle handling model. Comparison between the simulation results and the field test measurements demonstrates that the proposed tire model generates tire force more accurately than the conventional tire model, in which tire force is expressed by experimental polynomial equations.; Each 4WS control scheme is evaluated by the response of both J-turn maneuver and a lane change maneuver with a driver model. The driver model is an evaluation tool for vehicle handling performance used in the design stage. In the simulation of the J-turn, both control schemes reduced the yaw rate overshoot by 20% compared to that of a 2WS vehicle. Although the lateral deviations of a 4WS vehicle are almost the same as that of 2WS in the lane change maneuver, the yaw rate of 4WS was reduced by approximately 15% compared to a 2WS system. Moreover, the proposed control schemes are shown to be robust against uncertainties of vehicle parameters and external disturbances. The vehicle handling performance is significantly improved.