Optimisation of the die face design of high-speed progressive dies using machine learning algorithms

• Identification of correlations between FEM and real process data
• Development of explainable AI as well as human-in-the-loop approaches
• Optimisation of the active surface design of high-speed progressive tools

The central objective of the cooperation project is the identification of non-linear relationships between sensorically and numerically detectable process and state variables on the one hand and workpiece characteristics on the other in a multi-stage, high-speed forming process based on machine and deep learning methods and the resulting derivation of optimisation measures for their effective surfaces. In the development of AI models, multimodal approaches are to be tested for the first time, in which heterogeneous, experimental and simulative data are merged and used as input variables for machine and deep learning processes. Identified correlations or sensitivities between component properties and active surface parameters are then to be

subsequently be used to optimise active surface parameters. Both XAI and HITL approaches are to be tested and used in order to incorporate domain-specific knowledge into the modelling on the one hand and to test the plausibility of identified relationships on the other.

check the plausibility of recognised correlations and rule out spurious correlations. The geometric shape and dimensions of the cutting punch and die, the use of different dimensions of the sheet holder geometry, locally hardened punches and the design of the tribological system are considered to be possible optimisation options with regard to the effective surfaces.

The expected result of the research project is the development of a multimodal learning algorithm based identification of correlations between sensorically and numerically recorded process and state variables as well as product characteristics, the introduction of a procedure for the feedback of knowledge from the ML models by means of interactive learning techniques and the optimisation of the active surface design of high-speed progressive dies based on this.