Fighting the modeling bottleneck - learning models for production plants

Model-based approaches to system design, testing and diagnosis have been used successfully for more than 20 years. Looking at the large set of success stories, scientific papers, and commercial tools, one major critical point can be identified: The modeling bottleneck- modeling remains a tiresome, demanding task. Therefore, much work has been done to ease the manual modeling task. But there exists another approach to solve the modeling bottleneck: Learning models from empirical data. Model learning covers a large range of different algorithms: From rather simple model parametrization approaches to hard model synthesis questions. In this paper, new algorithms for the synthesis of states and transitions of timed automata are presented; algorithms are given for both discrete and hybrid systems. The application here are industrial plants and automation systems