Physics-informed Neural Networks for Elliptic Partial Differential Equations on 3D Manifolds

Abstract

Motivated by recent research on Physics-Informed Neural Networks (PINNs), we make the first attempt to introduce the PINNs for numerical simulation of the elliptic Partial Differential Equations (PDEs) on 3D manifolds. PINNs are one of the deep learning-based techniques. Based on the data and physical models, PINNs introduce the standard feedforward neural networks (NNs) to approximate the solutions to the PDE systems. By using automatic differentiation, the PDEs system could be explicitly encoded into NNs and consequently, the sum of mean squared residuals from PDEs could be minimized with respect to the NN parameters. In this study, the residual in the loss function could be constructed validly by using the automatic differentiation because of the relationship between the surface differential operators ∇ S /∆ S and the standard Euclidean differential operators ∇/∆. We first consider the unit sphere as surface to investigate the numerical accuracy and convergence of the PINNs with different training example sizes and the depth of the NNs. Another examples are provided with different complex manifolds to verify the robustness of the PINNs.

Physics-Informed Neural Networks for shell structures

“…We evaluate the accuracy of the solutions obtained with this framework in comparison to FEM solutions. Although we foresee this framework to be of primary interest to the computational mechanics community, this study may also, in a more general sense, shed light on the performance of PINNs in non-Euclidean domains for a set of non-trivial physical equations and extends prior studies (Tang and Fu, 2021). As the basis for our formulation, we consider the linear Naghdi shell model, suitable for the description of small-strain deformations of arbitrarily shaped shell structures.…”