Abstract
Keywords: Vehicle dynamics simulation, vehicle stability controllers, objective performance criteria, Design-of-Experiment, model-based optimization, probabilistic models, query, Active Learning
Closed-loop vehicle dynamics simulation is a well established tool both in function development, and in software and system testing for vehicle stability controllers. In this paper we describe how to use this tool in order to develop methods for automated performance calibration of controller sub-functions. The existence of validated simulation models, e.g. for vehicles and for tires, is of advantage for our purpose. Thus we expect this non-trivial problem to be solved at our starting point mainly by OEM. We define objective performance criteria for specific controller sub-functions that are active during certain driving maneuvers. The process of parameter tuning is then defined as a constrained higher dimensional optimization problem that is in general multi-objective. Based on powerful mathematical models and on optimization algorithms we can determine optimal controller parameters in sense of previously defined performance criteria. To ensure globally optimal parameters, we use a model-based optimization algorithm that benefits from methods of Pareto-Optimization, Design-of-Experiment, and Active Learning. We are confident that for some characteristic sub-functions our methods can also be transferred to in-car use for real vehicle testing.