Complete Model-Based Testing Applied to the Railway Domain

Abstract

Testing is the most important verification technique to assert the correctness of an embedded system. Model-based testing (MBT) is a popular approach that generates test cases from models automatically. For the verification of safety-critical systems, complete MBT strategies are most promising. Complete testing strategies can guarantee that all errors of a certain kind are revealed by the generated test suite, given that the system-under-test fulfils several hypotheses. This work presents a complete testing strategy which is based on equivalence class abstraction. Using this approach, reactive systems, with a potentially infinite input domain but finitely many internal states, can be abstracted to finite-state machines. This allows for the generation of finite test suites providing completeness. However, for a system-under-test, it is hard to prove the validity of the hypotheses which justify the completeness of the applied testing strategy. Therefore, we experimentally evaluate the fault-detection capabilities of our equivalence class testing strategy in this work. We use a novel mutation-analysis strategy which introduces artificial errors to a SystemC model to mimic typical HW/SW integration errors. We provide experimental results that show the adequacy of our approach considering case studies from the railway domain (i.e., a speed-monitoring function and an interlocking-system controller) and from the automotive domain (i.e., an airbag controller). Furthermore, we present extensions to the equivalence class testing strategy. We show that a combination with randomisation and boundary-value selection is able to significantly increase the probability to detect HW/SW integration errors.