Worst-case evaluation methods for vehicles and vehicle control systems.

Abstract

Worst-case evaluation methods are developed for the evaluation of dynamic systems in this dissertation. The objective of these methods is to systematically identify worst-case disturbances so that the performance of dynamic systems under extreme conditions can be evaluated. The generation of the worst-case disturbances is an optimization problem in a differential game framework. Depending on the number of players and the information structure, the worst-case evaluation problems can be classified into four types: one-player without preview information (1P), one-player with preview information (1PP), two-player without preview information (2P), and two-player with preview information (2PP). Classical optimal control and zero-sum two-player game theory are used to construct the worst-case disturbances. In general, the solution to a two-point boundary-value problem (TPBVP) is required for worst-case problems. When the system is linear, analytical solutions for the TPBVP can be obtained. For nonlinear systems, the worst-case problems need to be solved by numerically. Both analytical solutions and numerical methods are presented. The analytical solution of 2PP problems was derived, and is a key contribution of this dissertation. Two case studies on vehicle dynamic/control systems are presented to illustrate the procedures of the worst-case disturbance generation. The first case study is on the rollover and jackknifing of articulated trucks, which is formulated as a 1P problem. The second case study involves a vehicle dynamics control (VDC) system, whose worst-case disturbances are obtained by formulating a 2PP problem. In both case studies, the worst-case methods find the weakness of the target systems and result in instabilities. The identified worst-case disturbances do not exhibit features that can be easily constructed and explained by engineering intuition, which clearly shows the merits of the proposed methods.