Randomized reachability analysis in UPPAAL: fast error detection in timed systems

Abstract

Randomized reachability analysis is an efficient method for detection of safety violations. Due to the under-approximate nature of the method, it excels at quick falsification of models and can greatly improve the model-based development process: using lightweight randomized methods early in the development for the discovery of bugs, followed by expensive symbolic verification only at the very end. We show the scalability of our method on a number of timed automata and stopwatch automata models of varying sizes and origin. Among them, we revisit the schedulability problem from the Herschel–Planck industrial case study, where our new method finds the deadline violation three orders of magnitude faster: some cases could previously be analyzed by statistical model checking in 23 h and can now be checked in 23 s. Moreover, a deadline violation is discovered in a number of cases that were previously intractable. We have implemented the Randomized reachability analysis—and made it available—in the tool Uppaal. Finally, we provide an evaluation of the strengths and weaknesses of Random reachability analysis exploring exactly which types of model features hamper method’s efficiency.