Model-Based Testing Using Symbolic Animation and Machine Learning

Abstract

We present in this paper a technique based on symbolic animation of models that aims at producing model-based tests. In order to guide the animation of the model, we rely on the use of a deterministic finite automaton (DFA) of the model that is built using a well-known machine learning algorithm, that considers a complex model as a black-box component, whose behavior is inferred. Since the DFA obtained in this way may be an over-approximation and, thus, admit traces that were not admitted on the original model, this abstraction is refined using counter-examples made of unfeasible traces. The computation of counter-examples is performed using a systematic coverage of the DFA states and transitions, producing test sequences that are replayed on the model, providing either test cases for offline testing, or counter-examples that aim at refining the abstraction.