Data Science in Engineering

The future importance of machine learning and data analytics is indisputable, as a means of creating added value from data. In the special context of the Montanuniversität, the data emanates from physical, chemical and mechanical processes, which must obey the laws of physics. Consequently, a special branch of machine learning that incorporates models of these processes will be required. Such model based machine learning approaches allow to implement known dynamics or mechanics models. Furthermore, it enables the integration of dependencies, interactions and constraints which are essential for reliable, predictive and noise robust ML systems.

The challenge of establishing the successful use of machine learning in Leoben lies in the very nature of the problems we address in our research. Applying machine learning to observations and control of physical systems requires the embedded solution of inverse problems to ensure that the learnt results abide by the laws of physics. This is particularly challenging since it necessitates the combination of analytic methods and machine learning algorithms. Furthermore, in the fields of research addressed in Leoben, it is not sufficient to discover knowledge about a system, since we also need to establish understanding and to generate control signals. This requires the definition of appropriate representations for data, knowledge and understanding, together with mechanisms that connect various representations and algorithms running on different computational platforms.