Physics-Informed Neural Networks for Thermo-Responsive Hydrogel Swelling: Integrating Constitutive Models with Sparse...

Area of Science:

• Polymer Science
• Materials Science
• Computational Science

Background:

• Poly(N-isopropylacrylamide) (PNIPAM) hydrogels exhibit temperature-sensitive swelling, posing prediction challenges near the gel collapse temperature (GCT).
• Accurate modeling is crucial for applications in soft actuators and biomedical systems.

Purpose of the Study:

• To develop a physics-informed neural network (PINN) for accurate prediction of PNIPAM hydrogel swelling under varying conditions.
• To integrate governing physical equations with sparse experimental data for robust model development.

Main Methods:

• A PINN model incorporating a stabilized free-energy model was developed.
• Sparse data from free and uniaxially constrained swelling experiments were assimilated.
• The PINN was trained and validated across different crosslink densities.

Main Results:

• PINN significantly reduced test Root Mean Square Error (RMSE) by up to 65% and relative error under constraint by over 40%.
• Prediction intervals were narrowed without overfitting, improving coverage from 61.9% to 76.2% for free swelling.
• The model accurately captured transition slopes near GCT and thermodynamic stability.

Conclusions:

• PINNs provide a reliable surrogate model for thermo-responsive hydrogels, overcoming challenges in predicting swelling behavior.
• This framework enables accurate swelling and stress predictions with minimal data, facilitating the design of advanced hydrogel-based devices.