Using A multi-fidelity approach where data generated by a low-fidelity computational fluid dynamics (CFD) solution strategy is combined with physics-informed neural networks (PINN) to improve accuracy. First, a purely data-driven deep neural network is used to learn a nonlinear mapping from the input coordinates to the low fidelity data generated by CFD. Second, Transfer learning based on low-fidelity CFD data is used to initialize PINN without any high-fidelity training data. Additionally, a tradtional PINN simulation with random initialization was used to compare the convergence speed and accuracy of the two PINN approaches.
##########################################################
Codes and data used in the test cases presented in the paper:
Predicting high-fidelity multiphysics data from low-fidelity fluid flow and transport solvers using physics-informed neural networks.
##########################################################
Pytorch codes are included for the different test cases presented in the paper. Namely, 2D Lid driven cavity, 2D Fluid flow over a fin, 2D aneurysm, 2D Rotating heterogeneous porous medium ,2D Convection in a rotating homogeneous porous medium, and 2D multiphysics heat transfer in a fin.
##########################################################
Codes:
Codes for purely data driven and Multi-fidelity PINN are provided.
Note: Same codes are used for Multi-fidelity and traditional PINN. Set Flag-initialization = True and Flag-pretrained = False for traditional PINN (current codes are for Multi-fidelity PINN where Flag-initialization = False and Flag-pretrained = True).
###########################################################
Data:
The input data for all test cases are provided in the Data folder. CSV files were genearted by Fluent or FeniCS and represent low fidelity CFD results. All .pt files are generated using a purely data driven deep neural network to map input coordinates to the low-fidelity CFD data. The .pt files were used to initialize PINN.
###########################################################
Installation:
Install Pytorch:
https://pytorch.org/