Researchers have developed a novel polyconvex physics-augmented neural network (PANN) called PANN, designed to accurately and efficiently model the anisotropic hyperelastic response of representative volume elements (RVEs). This PANN construction utilizes learned paired directional features, which adapt to the RVE's specific behavior. The resulting energies inherently satisfy mechanical consistency properties like objectivity and polyconvexity, and have demonstrated superior performance in stress error and displacement error compared to unconstrained networks and reduced-order models, particularly beyond the training data. AI
IMPACT This research introduces a novel neural network architecture for complex material modeling, potentially improving simulations in computational physics.
RANK_REASON Academic paper detailing a new computational physics model. [lever_c_demoted from research: ic=1 ai=0.7]
- arXiv
- FE$^2$
- Ickån
- International Conference on Neural Networks
- PANN
- Sebastian Ares De Parga Regalado Dr.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
