Reproduce “Physics-Informed Neural Networks for Solving Contact Problems in Three Dimensions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tarik Sahin, Daniel Wolff, Alexander Popp
Researchers at the University of the Bundeswehr Munich developed an extended Physics-Informed Neural Network (PINN) framework to solve three-dimensional contact mechanics problems, leveraging a mixed-variable formulation and the Fischer-Burmeister function for inequality constraints. This approach achieved high accuracy against analytical solutions, with a data-enhanced PINN reducing L2 errors for normal stress in a Hertzian contact problem from 5.29% to 0.17%.
Key insights
- PINNs can effectively solve 3D contact mechanics problems by integrating a mixed-variable formulation, strategic hard/soft constraint enforcement, and a robust NCP function for KKT conditions.
- The Fischer-Burmeister function offers a numerically stable and differentiable method for enforcing complex inequality constraints within the PINN framework, applicable beyond contact mechanics.
- Data enhancement significantly improves the accuracy and robustness of PINN models, demonstrating the potential for hybrid modeling and the development of more reliable digital twins.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.