Differentiable Finite Element Framework for Learning Constitutive Models via Automatic Differentiation
FE-MAD (Finite Element-Based Material learning via Automatic Differentiation) is a novel end-to-end differentiable framework that integrates constitutive neural network models within a JAX-FEM nonlinear solver. It utilizes automatic differentiation for computing system gradients, eliminating the need for analytic adjoints or surrogate models. This approach robustly identifies model parameters from full-field deformation data and has been demonstrated on various datasets for incompressible isotropic hyperelasticity, showing generalization capabilities.