Scenario-Based Online Reachability Validation for CPS Fault Prediction

Unlike standalone embedded devices, behaviors of a cyber-physical system (CPS) are highly dynamic. Many parameter values (e.g., those related to nature environment and third party black box functions) are unknown offline. Furthermore, distributed sub-CPSs may exchange data online. In this article, we first propose the concept of parametric hybrid automata (PHA) to describe such complex CPSs. As some PHA parameter values are unknown until runtime, conventional offline model checking is infeasible. Instead, we propose to carry out PHA model checking online, as a fault prediction mechanism. However, this usage is challenged by the high time cost of state reachability verification, which is the conventional focus of model checking. To address this challenge, we propose that the model checking shall focus on online scenario reachability validation instead. Furthermore, we propose a mechanism to compose/decompose scenarios. Our scenario reachability validation can exploit linear programming to achieve polynomial time cost. Evaluations on a state-of-the-art train control system show that our approach can cut online model checking time cost from over 1 h to within 200 ms.