thomas-allard / datafrac

0 stars 0 forks source link

Welcome lol

This project investigates the applicability of physics informed neural networks (PINN) for discovering material behavior parameters for two conditions:

  1. elasticity
  2. linear elastic fracture

===============================================================================

Files and folders

===============================================================================

How to load a netcdf, indexing and slicing

Data for each time step is stored in a DataArray. A master DataSet is created by "stacking" all DataArrays for a simulations corresponding to the entire duration of the simulation. Each DataArray has a 2760 rows corresponding to the 2760 nodes of the mesh. The following fields are the corresponding columns of each data array:

- ux: displacement in the x-direction (mm). Nodal value
- uy: displacement in the y-direction (mm). Nodal value
- Mises: von Mises stress (MPa)
- S11: normal stress component in the x-direction (MPa)
- S22: normal stress component in the y-direction (MPa)
- S12: shear stress component (MPa)
- NE11: nominal normal strain in the x-direction (dimensionless)
- NE22: nominal normal strain in the y-direction (dimensionless)
- NE12: nominal shear strain (dimensionless)

Displacements, ux and uy, are nodal valued. All stress and strain components are normally elemental quantities and are calculated at the integration points. For the sake of this project, we will keep all quantities at the nodes, so these values are calculated at the nodes. This operation is conducted automatically by Abaqus and are the average stresses interpolated to the nodes. This operation is requested when building the simulation prior to submission in "build_model.py".

Currently, the DataSet is output as a NetCDF file, because it is easy. A saved DataSet may be loaded as follows: 'set = xr.open_dataset('{filename}.nc')