Model-based compositional verification approaches and tools development for cyber-physical systems

Ren, Hao
Major Professor
Ratnesh Kumar
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Journal Issue
Electrical and Computer Engineering

The model-based design for embedded real-time systems utilizes the veriable reusable components and proper architectures, to deal with the verification scalability problem caused by state-explosion. In this thesis, we address verification approaches for both low-level individual component correctness and high-level system correctness, which are equally important under this scheme. Three prototype tools are developed, implementing our approaches and algorithms accordingly.

For the component-level design-time verification, we developed a symbolic verifier, LhaVrf, for the reachability verification of concurrent linear hybrid systems (LHA). It is unique in translating a hybrid automaton into a transition system that preserves the discrete transition structure, possesses no continuous dynamics, and preserves reachability of discrete states. Afterward, model-checking is interleaved in the counterexample fragment based specification relaxation framework. We next present a simulation-based bounded-horizon reachability analysis approach for the reachability verification of systems modeled by hybrid automata (HA) on a run-time basis. This framework applies a dynamic, on-the-fly, repartition-based error propagation control method with the mild requirement of Lipschitz continuity on the continuous dynamics. The novel features allow state-triggered discrete jumps and provide eventually constant over-approximation error bound for incremental stable dynamics. The above approaches are implemented in our prototype verifier called HS3V. Once the component properties are established, the next thing is to establish the system-level properties through compositional verication. We present our work on the role and integration of quantier elimination (QE) for property composition and verication. In our approach, we derive in a single step, the strongest system property from the given component properties for both time-independent and time-dependent scenarios. The system initial condition can also be composed, which, alongside the strongest system property, are used to verify a postulated system property through induction. The above approaches are implemented in our prototype tool called ReLIC.