Master Thesis Development of New Data-Driven Method for Metal Fatigue Assessment

In recent years, neural networks / machine learning methods have attracted increasing attention due to their ability to model complex, non-linear interactions between given input and output variables. On the one hand, this property is a great strength. On the other hand, the training requires rather large data sets and the interpretability due to the black-box character is an obstacle for many engineering problems, where small data sets and physical relationships or boundary conditions play an essential role. In engineering applications, interpretable analytical equations are preferred not only for trustworthiness but also for understanding interpolations and extrapolations given the limited and heterogeneous data. Symbolic regression method is a data-driven approach for learning analytical equations. Implementations are currently limited and research in this field is ongoing.

• Therefore, you will contribute to the field of new explainable data-driven models and enhance their usage in engineering applications.
• You will develop the new data-driven methodology and implement it in a Python-based toolbox. Assumptions, limitations and boundary conditions of the new method should be investigated on artificial data.
• For a real-world scenario, the methodology is tested on material fatigue assessment data, and you should compare it to a state-of-the art fatigue assessment method for reliability.
• Big challenge: Robust training by ensuring stable numerical optimizations.

• Education: Master studies in the field of Mathematics, Computer Science, Engineering or comparable
• Experience and Knowledge: strong mathematical background; strong programming skills (preferably in Python); knowledge of numerical optimization is a plus
• Personality and Working Practice: you are able to communicate your ideas clearly, organize your tasks efficiently and take responsibility for them; you work effectively with others and maintain focus on team objectives
• Languages: very good in German or English