Friction models for cold forging based on machine learning

• Applicant: Professor Dr.-Ing. Peter Groche
• Subject Area: Primary Shaping and Reshaping Technology, Additive Manufacturing
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 563455474

In cold forging, the friction conditions have a significant influence on the accuracy of the manufactured components and process reliability. Precise friction modelling is therefore crucial for the design of the forming processes. The tribology of cold forging has a number of special features: The highest tribological loads (in particular contact normal stresses and surface enlargements) in forming technology typically occur in cold forging processes. In addition, the tribological system is dependent on a large number of non-linear influencing variables and is therefore challenging to model. The mathematical description of friction has so far mainly been based on "grey box" models such as Coulomb's friction law. These are convincing due to their simplicity and user-friendliness, but cannot fully represent the real complexity of the tribological system. This project investigates "black box" models based on machine learning (ML) as a possible solution for friction modelling of the complex tribological system in cold forging. The structure of the ML models makes it possible to take into account a larger number of influencing variables on the friction behaviour of the tribological system and to model non-linear dependencies between the influencing variables better than existing friction models. In addition, ML methods open up the possibility of using large amounts of data from direct tribometer tests in the form of time series and thus increase the information content of the database compared to conventional evaluation methods for model tests. The integration of the models trained in the project into the FE simulation improves the accuracy of the friction force calculation and thus enables important design decisions to be made as part of the process design, such as determining the shape of the active surfaces for a target-oriented material flow and identifying suitable measures for robust process control. The main objectives of the project are: to develop a method for creating ML-based friction models on the basis of time series from non-stationary tribometer tests. The validation of the methodology is based on the comparison of experimentally and numerically determined force curves and the workpiece geometry.

• DFG Programme: Research Grants