Autonomous Generation of data-driven Features for Parametric Surface Models in Reverse Engineering

• Applicant: Dr.-Ing. Stefan Holtzhausen
• Subject Area: Engineering Design, Machine Elements, Product Development
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 544804370

Due to the need for advanced geometry analysis processes in the product development of engineering parts, the approach of reverse engineering (RE) is increasingly drawing attention. However, the current state of research on RE automation needs to be improved due to the non-continuous computer-based recognition of geometric features, which currently relies on cooperative human-machine interaction. Therefore, more advanced computer-based recognition capabilities are required to increase the resilience of geometry analysis processes in RE. The aim is to enhance the robustness of computer-based automation against variable acquisition data qualities and model complexities of engineering parts so that the same reconstruction quality in the RE process can be achieved with less or no human recognition capabilities. Nowadays, the research focus shifts to deep learning (DL) techniques, which allow robust handling of geometric deviations from data-driven fuzziness and logical inconsistencies. Therefore, the research project's objective is to integrate novel DL approaches into RE processing, particularly into the recognition activities of the geometric features of engineering parts. Essentially, it needs to be resolved how the reverse engineering process chain can be extended by DL methods so that geometry features and geometric constraints are autonomously identified. Furthermore, a unified detection of the global constraints is intended to provide the needed information of modeling operations to make the RE process more robust and support the derivation of a design sequence. In this context, the DL probability statements are supposed to enable the continuous classification of subsurfaces based on geometric features and the distinction of different subsurface instances. As a result, a consistent methodology and continuous DL-based geometry reconstruction of 3D scans will be established, autonomously generating data-driven surface reconstruction features without mandatory expert intervention. This could enable much more productive use cases in the product development process in the future.

• DFG Programme: Research Grants
• Co-Investigator: Professorin Dr.-Ing. Kristin Paetzold-Byhain