Learning Timed Automata from Interaction Traces

摘要

The design of load-critical human-machine systems presumes thorough modelling and analysis of interaction profiles the systems are meant to withstand at peak loads. The need for mathematical modelling of interactions is often ignored due to significant modelling effort and lack of relevant tools. We propose an algorithm for automatic learning a subclass of Uppaal timed automata models from system and its environment interaction logs. The learning method relies on synchronous communication assumption that is characteristic to communication protocols of networked HMS distributed components. The method is demonstrated on IEEE1394 protocol learning example. Beside enhancing automatic test generation, the learned model allows verifying test feasibility and test optimization already in early phases of test design.