Fundamental investigations regarding the automated design of microfinishing for rotationally symmetrical components to machine functional surfaces using machine learning methods

• Applicants: Professor Dr.-Ing. Dirk Biermann; Professor Dr.-Ing. Bernd Kuhlenkötter
• Subject Area: Production Systems, Operations Management, Quality Management and Factory Planning ; Automation, Mechatronics, Control Systems, Intelligent Technical Systems, Robotics
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 570592100

The proposal addresses the utilization of artificial intelligence for designing the process of the microfinishing with the aim of reducing the consumption of human and material resources. Suitable process strategy developments in microfinishing are currently limited to the experience of machine and tool manufacturers and experienced machine operators. However, the number of experienced skilled workers is gradually decreasing due to various influences, such as demographic change. In addition, microfinishing usually involves a multi-stage (often three-stages) process, whereas the stages must be aligned to each other. Therefore, test trials are required. The digitalization and further development of AI offers the opportunity to compensate for the loss of knowledge and to make empirical knowledge permanently available and to make process predictions to reduce the required test trials. Microfinishing is a precision honing process whose complexity limits the use of simulation techniques and analytical descriptions. For this reason, the application of AI, in particular machine learning (ML), is a suitable method for extracting process knowledge and describing high-dimensional, non-linear dependencies. The aim is to train ML models on the basis of experimental data that needs to be generated. Initially it is important to investigate, if the same process parameters lead to comparable final results of the surface topography. This needs to be taken into account for the database generation. From a ML model of the first processing stage of microfinishing that will be developed, it is examined to what extent the ML models for the two remaining tool stages can be created by transfer learning. The aim is to reduce the required data and thus reduce the experimental effort. At the same time, this ensures generalization to tool combinations not included in the project with less additional experimental effort. Finally, the representative process models (surrogate models) are used as optimization objects in order to achieve an intelligent, adaptive and optimized process design. The optimization is preferably carried out using reinforcement learning or genetic algorithms, so that optimal parameters are predicted for each processing stage over the entire microfinishing process to achieve an optimal end result (after the third stage). In addition to the process design, suitable measurement technology is used in the experiments, to determine to which extent the interaction between the material removal rate and tool wear can be described and how this can be used for the ML. Moreover, the aim is to investigate the dependency between the initial topography and final topography, taking into account the process input parameters. After a successful optimization with validation, a final recommendation model for dealing with a wide range of input and influencing variables will be developed.

• DFG Programme: Research Grants