 |
NN-PGD for surrogate modeling of PDEs on parametrized domains
1 : Laboratoire de mécanique des solides, CNRS, École polytechnique, Institut Polytechnique de Paris
Polytechnique - X
2 : Laboratoire de mécanique des solides, CNRS, École polytechnique, Institut Polytechnique de Paris
* : Auteur correspondant
Polytechnique - X
This work presents an extension of the Neural Network–Proper Generalised Decomposition (NN-PGD) framework for constructing surrogate models of PDEs defined on parametrized domains. Using a mapping onto a reference domain, the method allows a single model to provide solutions across a range of geometries. The framework combines the PGD with physics-informed training, enabling the modes to be learned directly from the governing equations. The approach is demonstrated on a 2D linear elasticity problem on a parametrized hexagonal domain.
