Digital materials created by multi-material 3D printing are modeled as mixtures of stiff and compliant components, resulting in a wide range of stiffness and complex nonlinear behaviors. Classical finite-strain viscoelastic models use strain energy functions and internal variable evolution but may lack flexibility across different materials and loading conditions.
The proposed data-driven framework extends a classical formulation by Bergström and Boyce, maintaining multiplicative kinematics, invariant-based strain-energy functions, and a scalar dissipative evolution law. It predicts model parameters directly from composition or constructs polyconvex strain-energy functions using neural ordinary differential equations (NODEs). The kinetics of nonequilibrium behavior are learned either through direct parameter identification or constrained neural networks.
Using multi-rate uniaxial compression data across various compositions, the model captures rate-dependent stiffness and hysteresis while ensuring thermodynamic consistency. This approach offers a flexible, generalizable way to model complex behaviors in digital materials, aiding engineers in simulation and material design.
Source: https://arxiv.org/abs/2609.04541