Reproduce “Physics-informed neural network for modeling dynamic linear elasticity” 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
Vijay Kag, Venkatesh Gopinath
Kag and Gopinath at the Bosch Research and Technology Center developed Physics-informed Neural Networks (PINNs) for dynamic linear elasticity, capable of predicting full-field displacements and stresses from sparse data. Their PINN-based surrogate models achieved NRMSE errors in the order of 10 -3 to 10 -2 for unseen material parameters in 2D and 3D problems, reducing material identification computation time from 20 minutes (FEM) to seconds.
Key insights
- PINNs can accurately predict full-field displacements and stresses in dynamic linear elasticity from very sparse boundary data, achieving NRMSE errors in the order of 10 -3 to 10 -4 for forward problems.
- PINN-based surrogate models for material property estimation drastically reduce computation time from approximately 20 minutes (FEM) to seconds for predicting solutions with new material parameters in both 2D and 3D.
- For inverse problems, accurate identification of unknown material parameters like Lamé constants requires careful selection of activation functions (e.g., sigmoid) for parameter mapping and proper treatment of boundary conditions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.