Download PDFOpen PDF in browserComparison of Conventional and Data-Driven Friction Models for Cold Forging on a Shear Stress Level8 pages•Published: August 6, 2026AbstractFriction modelling in finite element simulations is largely based on state-of-the-art analytical approaches, ranging from the Coulomb friction model and the shear friction model to advanced friction models that incorporate multiple influencing variables to represent the fluctuating contact conditions. This work presents a systematic benchmark of such state-of-the-art friction models, complemented by data-driven machine-learning-based approaches. A tribometer-based benchmark using sliding compression tests is conducted under identical boundary conditions and over a wide range of tribological loads. To enable a physically consistent comparison of fundamentally different friction formulations, model performance is specifically evaluated on the level of frictional shear stresses. The results show that the neural-network-based friction model consistently achieves the highest predictive accuracy, exhibiting the lowest prediction error. In comparison with the Coulomb model, which utilises standard constant values, the mean absolute error (MAE) is reduced from 51.38 MPa to 9.35 MPa. This equates to an error reduction of 81.8%.Keyphrases: cold forging, data based friction modelling, tribometrics In: Numpon Mahayotsanun (editor). Proceedings of The 11th International Conference on Tribology in Manufacturing Processes & Advanced Surface Engineering, vol 4, pages 49-56.
|
