30 % of the development costs of a tool in sheet metal forming are caused by the manual tool try out process. This process is complex, very time-consuming, and physically demanding. It is a challenging task due to complex interactions and high uncertainties, and requires an individual approach for each tool.
Essentially, two types of degrees of freedom can be influenced: (i) the active surfaces of the forming tool and (ii) the adjustable parameters of the “toolmachine” system. Starting from a formal description of the final product to be manufactured, the forming tool is developed. Ideal values of the active surfaces and machine parameters are defined during die development (design function f). The tool commissioning including the mechanical try out is then carried out on a tryout press. A successful optimization of the degrees of freedom (improvement function g) by shop workers enables the release to the production press.
Currently, functions f and g are implemented manually by human experts, relying heavily on experience and heuristic knowledge. This leads to non-deterministic results, extended start-up phases, and suboptimal products. In this project, machine learning methods that learn f and g from data will be developed to both automate the tool try out and adapt the design of the tool to the complex interactions of the real system already in the tool development process. On this basis, the following three research questions are addressed in this project:
- What kind of neural networks with which network topology are suitable for the automated generation of the active surfaces of the forming tool, i.e., how can f be learned from data to solve the design problem of the active surfaces.
- What kind and topology of neural networks is suitable for learning the function f in terms of machine parameters such as force and velocity?
- How can the human operator be assisted in the task of handling the tryout press? I.e. can we use data and technical know-how to calculate function g and automate the activity?
Duration: 1.7.2023 to 1.7.2026 (Phase I only)
This project is funded by the DFG-Priority Programm „Datengetriebene Prozessmodellier
Letzte Änderung: 15. March 2023