Finite Element-Based Material Learning via Automatic Differentiation: Learning constitutive neural network models from full-field deformation data
Researchers have developed FE-MAD, a novel framework that integrates constitutive neural networks with a JAX-FEM nonlinear solver. This end-to-end differentiable system uses automatic differentiation to compute gradients, enabling efficient learning of material models from full-field deformation data. The framework was successfully demonstrated on hyperelasticity problems, identifying material properties from various experimental datasets and generalizing to unseen samples. AI
IMPACT This framework could accelerate the discovery and design of new materials by enabling more efficient learning of complex constitutive models from experimental data.