Optimal Smooth Switching of Controllers for Heavy-Duty Vehicles

Abstract

Abstract Heavy road vehicles currently contain different dynamics control systems such as advanced driver assistance control systems (ADAS) and simple rule based control systems. As the transition towards full electrification of heavy vehicles and fully autonomous vehicles continues, newer control systems are being added to current ones. Thus, there is a need to be able to switch between different vehicle control systems. The process of making these types of switches smooth is the topic of this thesis. Smooth switching is considered to occur when there are no large bumps or oscillations during the switching process. Research done in the area of switching systems usually focuses on having the outputs of the controller coming online to be close to the outputs of the controller going offline so that the transition is smooth. Trying to match the controller outputs does not work for heavy vehicles because of the number of different actuators and the differences in the response rates of these actuators. The contribution of this thesis is to develop and analyze methods of switching between controllers that may have different goals and are acting on differing actuators. Switching methods of different complexities are developed and analyzed. Preparing the offline controller by connecting it to a model of the real vehicle is shown to decrease the dynamics that occur when the new controller is connected. Using linear interpolation and sigmoid functions to switch from the outputs of the initial controller to the final controller are tested using a realistic simulation environment developed by Volvo Trucks. Filtering these interpolation functions depending on the actuator response rates shows significant decrease in the disturbance to the vehicle during the switching process. Switching using model predictive control as an in between controller to switch from the tire force forces of the initial controller to the tire forces of the final controller is shown to be feasible. This method has the advantage of achieving optimal switching depending on the priorities of the control designer. Another advantage of the model predictive controller is obtained if the timing of the switch is known before it happens. Using this preview information, the actuators can start changing their outputs even before the switch time for optimal switching performance. Since the model predictive controller requires knowledge of all of the system states a Cubature Kalman filter is developed. This Kalman filter also estimates the tire stiffnesses because it is relatively difficult to know the tire properties of a truck.