9:30 AM - 9:45 AM
[21001-05-05] Inverse Analysis of Heat Equation using Variational PINN (Proceedings of Symposium on Applied Mechanics)
Keywords:Machine Learning, Physics-Informed Neural Network, Inverse Analysis
PINN (Physics-Informed Neural Network) has been one of the most promising machine learning models to evaluate physical phenomena. However, several scenarios have been found, where PINN fails to capture the solutions. One of the potential reasons is that, as is characterized to be a collocation method to evaluate the governing PDEs, it struggles to recover the entire solution when the solution includes a steep change. In this context, hp-VPINN is a powerful candidate to mitigate this issue since it can refine the mesh / polynomials progressively and locally. We investigate its effectiveness with particular focus on the inverse analysis application.