Modeling elastoviscoplastic materials using physics-informed neural networks.

Area of Science:

• Materials Science
• Rheology
• Computational Mechanics

Background:

• Elastoviscoplastic (EVP) materials exhibit complex nonlinear rheological behavior, challenging traditional predictive modeling, especially under large amplitude oscillatory shear (LAOS).
• Existing models often suffer from computational complexity, poor generalizability, and difficulties fitting noisy experimental data, limiting their practical application.

Purpose of the Study:

• To re-evaluate rheological modeling of EVP materials using a novel physics-informed neural network (PINN) approach.
• To develop a data-efficient, differentiable, and generalizable method for predicting nonlinear viscoelastic behavior.

Main Methods:

• Embedding a modified Saramito model within a PINN framework to directly fit time-dependent stress data.
• Incorporating a shear-thinning formulation to account for decreasing viscosity at higher strain amplitudes.
• Validating the PINN approach using both synthetic and experimental data.

Main Results:

• The PINN approach demonstrated stable recovery of physical parameters from noisy data, circumventing the need for gradient estimation.
• The method showed enhanced interpretability and improved predictive capabilities across various strain amplitudes.
• Successful validation using both synthetic and experimental datasets confirmed the model's robustness.

Conclusions:

• The proposed PINN framework offers a powerful and data-efficient alternative for rheological modeling of EVP materials.
• This approach effectively bridges microstructural deformation modes with macroscopic rheological predictions.
• The study provides a versatile tool for understanding and predicting nonlinear viscoelastic behavior in soft matter systems.