Skip to content

End-to-end pipeline for training a convolutional network directly with 2D/3D simulations, and visualising regressed results.

Notifications You must be signed in to change notification settings

amorx1/HFM-Pipeline

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

39 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

HFM-Pipeline

Complete Python pipeline for:

  1. Extracting spatio-temporal data from 2D/3D Hidden Fluid Mechanics simulations (.vtu data)
  2. Creating, training and testing "Physics-Informed" convolutional neural network, with manual hyperparameter tuning
  3. Visualizing regressed spatio-temporal mesh data

The program is a solution for the extensive data manipulation that needed to be done when moving data to and from a neural network pipeline, as part of a research project. The data was provided in the form of 2D CFD simulations, from which spatial and temporal coordinates, and concentration, velocity and pressure field data had to be retrieved and exported to csv files. These then had to be iterated through to extract and append the data to another csv file for each timestep, before converting everything again to a .mat file as the input to the the neural network.

The neural network was configured to output predictions in the form of .mat files, meaning these had to be converted back into csv and then vtu files in order to visualize the results. This could be done in MATLAB, however, Python > MATLAB. Intended functionality aims to provide the option of either outputting vtu data for visualization in ParaView, or directly visualizing in Python IDLE.

About

End-to-end pipeline for training a convolutional network directly with 2D/3D simulations, and visualising regressed results.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages