Surrogate-based prediction of cycle fatigue strength of rotating shafts and process optimization using flexible Kriging models

• Applicant: Professor Dr. Roland Herzog
• Subject Area: Mathematics ; Metal-Cutting and Abrasive Manufacturing Engineering
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 547638154

Ensuring the functional properties of industrially manufactured components is an important factor in today's manufacturing technology. This usually requires time-consuming and costly measurement campaigns. Therefore, the main objective of this project as part of a package application is therefore to develop models that enable reliable predictions of the functional properties of individual component based solely on data that can be collected during the manufacturing process. The cycle fatigue strength of rotating shafts from martensitic steel that have been produced from an initial heat treatment and subsequent form turning is considered as an example. In this context, the research subject of the proposed project are methods of supervised machine learning that are able to predict such functional properties. They are utilizing a large number of heterogeneous data that is collected before and during the production process, such as material data, component dimensions and, for instance, speed and feed during milling. Specifically, so-called Kriging models will be used, also known under the name of Gaussian process regression. These are not only the central tool for creating the data-driven process-to-property models that are characteristic to the research unit, but will also be a direct object of research due to the specific requirements of the applications under consideration. For example, problem-specific regression and correlation functions must be developed, taking into account the heterogeneity of potential influencing variables, as well as fast iterative solution methods for estimating the associated parameters. Further research is required due to possibly incomplete data sets and the question of which process variables actually have a significant influence on the cycle fatigue strength of the manufactured components.

• DFG Programme: Research Grants