Physics-informed data-driven discovery of constitutive models with application to strain-rate-sensitive soft...

Area of Science:

• * Computational Mechanics
• * Materials Science
• * Machine Learning

Background:

• * Developing accurate constitutive models for strain-rate-sensitive soft materials remains a challenge.
• * Traditional models struggle to capture complex behaviors and require extensive experimental data.
• * Integrating physics principles with data-driven methods offers a promising avenue for improved modeling.

Purpose of the Study:

• * To propose a novel data-driven constitutive modeling approach combining continuum thermodynamics and machine learning.
• * To demonstrate the model's efficacy on strain-rate-sensitive soft materials.
• * To enforce physics-based constraints within the machine learning framework.

Main Methods:

• * Viscous dissipation-based visco-hyperelasticity framework with stress decomposition.
• * Irreducible integrity basis for stress component representation.
• * Gaussian process regression surrogate models trained on strain and strain rate invariants.

Main Results:

• * The physics-informed data-driven model accurately captures stress-strain-strain rate responses.
• * Improved prediction accuracy and generalizability across multiple deformation modes were achieved.
• * The model demonstrates compatibility with limited data sets.

Conclusions:

• * The proposed approach successfully integrates physics-based constraints into a data-driven constitutive model.
• * This method offers a more robust and generalizable alternative to classical and purely data-driven models.
• * The framework shows significant potential for modeling complex material behaviors with reduced data requirements.