Enhanced physics-informed neural networks for hyperelasticity

Abstract

Physics-informed neural networks have gained growing interest. Specifically, they are used to solve partial differential equations governing several physical phenomena. However, physics-informed neural network models suffer from several issues and can fail to provide accurate solutions in many scenarios. We discuss a few of these challenges and the techniques, such as the use of Fourier transform, that can be used to resolve these issues. This paper proposes and develops a physicsinformed neural network model that combines the residuals of the strong form and the potential energy, yielding many loss terms contributing to the definition of the loss function to be minimized. Hence, we propose using the coefficient of variation weighting scheme to dynamically and adaptively assign the weight for each loss term in the loss function. The developed PINN model is standalone and meshfree. In other words, it can accurately capture the mechanical response without requiring any labeled data. Although the framework can be used for many solid mechanics problems, we focus on three-dimensional (3D) hyperelasticity, where we consider two hyperelastic models. Once the model is trained, the response can be obtained almost instantly at any point in the physical domain, given its spatial coordinates. We demonstrate the framework's performance by solving different problems with various boundary conditions. KeywordsComputational mechanics • Curriculum learning • Fourier transform • Meshfree method • Multi-loss weighting • Partial differential equations Recently, rapid growth in the utilization of deep learning (DL) and data-driven modeling has been seen in computational solid mechanics [4

Stiff-PDEs and Physics-Informed Neural Networks

“…Recently, physics-informed neural networks (PINNs) have gained popularity due to the novel approach for solving forward [1][2][3][4] and inverse problems [5][6][7] involving PDEs using neural networks (NNs). Unlike conventional numerical techniques for solving PDEs, PINNs are non-data-driven meshless models that satisfy the prescribed initial (IC) and Llion Evans, Michelle Tindall, and Perumal Nithiarasu have contributed equally to this work.…”

Improving the accuracy of the deep energy method

“…The modification allowed the successful resolution of stresses around the stress concentration regions. Abueidda et al 39 enhanced the model by developing PINN that combines residuals of the strong form and the system's potential energy. The proposed formulation yielded a loss function with many loss terms.…”

On the use of graph neural networks and shape‐function‐based gradient computation in the deep energy method

“…Besides, researchers have used a mixed formulation, combining both the energy method and the strong form, to capture the mechanical response with high solution gradients and stress concentrations. [24][25][26] In almost all cases, these spatial gradients are obtained utilizing automatic differentiation (AD) of the NN model. [27][28][29] This approach is widely used since AD is already applied in the backpropagation step during NN model training.…”