On September 7, 2026, a research preprint (arXiv:2609.04541) introduced a data-driven constitutive modeling framework designed to accurately simulate the mechanical response of digital materials fabricated via multi-material 3D printing. The approach overcomes major hurdles in predicting the stiffness and energy dissipation of hybrid materials whose stiffness can vary by more than an order of magnitude depending on constituent mixing ratios.
According to the paper, digital materials are manufactured through controlled blending of rigid and flexible constituents. Consequently, they exhibit pronounced nonlinear hyperelastic and viscoelastic behavior that varies simultaneously with material composition and strain rate. Traditional finite-strain analytical models typically represent this behavior using fixed, closed-form strain energy functions for both equilibrium and non-equilibrium stresses, leading to limited flexibility when generalizing across diverse material formulations or varying loading rates.
Physics-Informed Machine Learning Architecture
To address these limitations, the research extended the classical Bergström-Boyce model. The new framework preserves fundamental physical structures of continuum mechanics, including multiplicative kinematics, invariant-based strain energy functions, and a scalar dissipation evolution law driven along normalized non-equilibrium deviatoric stress.
The core innovation lies in the integration of machine learning:
- Equilibrium Branch: The system can directly predict closed-form model parameters based on composition or autonomously learn polyconvex strain energy functions using Neural Ordinary Differential Equations (Neural ODEs / NODEs). - Non-Equilibrium Dynamic Branch: The training process similarly determines closed-form parameters or applies physics-constrained neural networks.
Validation on multi-rate uniaxial compression datasets demonstrated that the model accurately captures hysteresis loops and rate-dependent stiffness across broad composition ranges while strictly maintaining thermodynamic consistency.
Current Limitations and Outlook
Currently, the report provides experimental validation based solely on laboratory uniaxial compression test data. The authors have not yet released an open-source codebase or evaluated computational performance when embedding the model into commercial finite element analysis (FEA) software under complex multiaxial deformation states.