Researchers have developed a novel variational physics-informed graph neural network (PI-GNN) designed to more accurately model stress localization in heterogeneous solids. Unlike traditional physics-informed neural networks (PINNs) that rely on penalty terms or specific regularization widths, this PI-GNN integrates heterogeneity directly into the mesh graph and minimizes total potential energy. This approach results in significantly reduced errors in stress and displacement calculations, particularly across wide ranges of material stiffness mismatches, outperforming strong-form PINNs and achieving results comparable to finite element methods. AI
IMPACT This PI-GNN approach could enhance the accuracy and efficiency of simulations in solid mechanics, potentially impacting fields like materials science and engineering design.
RANK_REASON Academic paper detailing a new method in computational mechanics. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- finite element
- Gotit.pub
- heterogeneous solids
- Hugging Face
- Neo-Hookean hyperelasticity
- physics-informed neural networks
- PI-GNN
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
