In case of safety critical systems, it is extremely important to ensure fault-free operation, for example, by using model-based technologies. These systems are typically real-time systems where it is particularly important to examine the effect of timing parameters, as design problems often arise from complex timing conditions. However, in many cases the internal implementation of some components of the system is unknown (for example, in case of third-party components) and therefore there is no formal model or specification of its operation.
Automaton learning algorithms can be used to automatically derive a finite automaton-based behavioral model of the system, based on the observed operations. Real-time systems, however, raise new challenges to automaton learning, as the recognition and learning of time-dependent behaviors become rather important.
In my work I study learning algorithms for timed systems. I have implemented a software, based on a graph database, that allows the analysis of the time-dependent behavior of systems using a real-time automaton learning algorithm. I have also examined possible modifications and improvements of the algorithm.
The software I have implemented, supports the analysis of system components that was previously impossible due to their time-dependent behavior. Based on the resulting specification, the behavior of the system can be understood, analyzed, anomaly-detecting monitors can be derived, and documentation and tests can be generated.
The goal of my thesis is to present the software I have implemented and the algorithm it uses as well as the proposed improvements. The implementation details are also presented. The efficiency of the improvements is demonstrated by measurements.