In the era of artificial intelligence (AI), display products such as televisions, computers, and foldable smartphones serve as critical interfaces, benefiting significantly from the application of soft adhesive materials due to their exceptional performance. The mechanical characteristics of these adhesives are essential for the design and reliability assessment of display products. Constitutive models, which mathematically depict the material response under various physical conditions, are crucial for the accurate simulation of soft adhesives, yet modeling these materials poses significant challenges. Traditional constitutive modeling approaches have been either predominantly physics-based or entirely data-driven. While physics-based models provide clear insights into material behavio...[ Read more ]

In the era of artificial intelligence (AI), display products such as televisions, computers, and foldable smartphones serve as critical interfaces, benefiting significantly from the application of soft adhesive materials due to their exceptional performance. The mechanical characteristics of these adhesives are essential for the design and reliability assessment of display products. Constitutive models, which mathematically depict the material response under various physical conditions, are crucial for the accurate simulation of soft adhesives, yet modeling these materials poses significant challenges. Traditional constitutive modeling approaches have been either predominantly physics-based or entirely data-driven. While physics-based models provide clear insights into material behavior, they are limited by the complexity of model selection and rigidity of mathematical formulations. On the other hand, data-driven models, although highly flexible and precise with adequate data, necessitate extensive datasets for accurate prediction and generalization. This research introduces a hybrid constitutive modeling strategy that synergizes the strengths of both methodologies to address their inherent limitations.

We propose an innovative constitutive modeling technique for soft adhesives utilizing a physics-informed neural network. This method incorporates foundational knowledge from material science and mechanics into neural networks, yielding a physics-informed, data-driven constitutive model. The integrated principles include thermodynamic consistency, conditions for increasing strain energy, the non-negativity of strain energy, energy normalization, and the principles of symmetry and objectivity in mechanics. This technique enables automated, high-precision modeling with minimal experimental data on novel materials.

This work seeks to harness the synergies of physics and AI by merging them into a physics-informed neural network for soft adhesive materials. Achieving automated, high-precision modeling with minimal manual intervention allows for accurate representation of the complex responses of soft adhesives. Ultimately, this framework streamlines the constitutive modeling of new materials, expedites material selection, and accelerates material development.