Solving Modeling Problems with Machine Learning -- A Classification Scheme of Model Learning Approaches for Technical Systems

The manual creation and maintenance of appropriate behavior models is a key problem of model-based development for technical systems such as vehicles or production systems. Currently, two main approaches are considered promising to overcome this bottleneck: (a) The fields of software and system engineering try to develop better methods and tools to synthesize models manually. (b) The field of machine learning tries to learn such models automatically, by analyzing observations from running systems. While software and system engineering approaches are suited for new systems for which modeling expertise is at hand, machine learning approaches are suited for the handling of running systems such as monitoring, diagnosis, and system optimization. Given this background our contributions are as follows: To organize existing research, Section 2 introduces a classification scheme of models and model learning algorithms. Section 3 reports on two different case studies to illustrate the power of the machine learning approach.