Pinned Grain Growth Progress Report III

Posted by Fred Hohman, David Montes de Oca Zapiain on October 28, 2015

This is the post equivalent of the presentation given for our third progress report.

Outline

  • Current Status
  • Preliminary I/O for our surrogate model
  • Analysis of PCA results

Current Simulations Performed

Picture

I/O

Picture

Chord Length Detailed

Picture

Picture

Picture

Expansion of our Workflow

Picture

Some Preliminary PCA Results

  • 1 case (specific pin shape, pin percentage) for 10 time-steps
  • Can already see converging to steady state by clustering of data points
  • Cross correlation of boundary and pins

Picture

  • 3 cases varying pin shape for given percentage for 5 time-steps
  • Can see path in PCA space as microstructure evolves

Picture

Picture

So, Our Input PCA Data Matrix…

  • Remember our data matrix from presentation II?
  • Size: (n_simulations, 300^3) => (184, 27000000)
  • Main issue in 3D 2pt statistics of 300^3 microstructure computation is memory

So, Our Input PCA Data Matrix…

  • That’s big (really big)
  • We need a better way to work with this
    • Suggestions welcomed
    • Planning to talk to Ahmet about recent reduction work
    • 2pt statistics in 2D tappers off, expect similar feature in 3D

Attempt of PCA on Chord Length Distribution

  • Example of what not to do
  • PCA on chord length distribution without normalization

Picture

Next Steps

  1. Determine appropriate truncation of microstructure data to perform PCA without the need for large amounts of memory
    • Talking to Ahmet
  2. Gather all chord length distribution data
    • Done as of this morning!
  3. Run PCA on normalized chord length data
  4. Work with David Brough to use latest fitting code to create linkages