Yufeng-shen/StrainRecon

pycuda code for strain reconstruction

StrainRecon

Code for the papar: Voxel-based strain tensors from near-field High Energy Diffraction Microscopy

@article{SHEN2020100852,
title = {Voxel-based strain tensors from near-field High Energy Diffraction Microscopy},
journal = {Current Opinion in Solid State and Materials Science},
volume = {24},
number = {4},
pages = {100852},
year = {2020},
issn = {1359-0286},
doi = {https://doi.org/10.1016/j.cossms.2020.100852},
url = {https://www.sciencedirect.com/science/article/pii/S1359028620300504},
author = {Yu-Feng Shen and He Liu and Robert M. Suter},
keywords = {Microstructure, Metals, Strain, nf-HEDM, 3DXRD}
}

This software is an intra-granular strain tensor reconstruction toolkit for near-field high-energy X-ray diffraction microscopy (nf-HEDM). This readme file contains information about its Usage, its Dependencies on other libraries, the Structure of the package, the Formats of its input and output files.

Dependencies

	version
CUDA	9.1
python	3.6.6
pycuda	2018.1.1
numpy	1.15.3
scipy	1.1.0
matplotlib	3.0.0
PyYAML	3.13
h5py
jupyter	1.0.0

Usage

Step 1: Peak File Generation

• Case 1: Synthetic Bragg peak patterns

  1. Generate the synthetic sample, e.g. micFile/Ti7FFT.hdf5

  2. Write the configure file, e.g. ConfigFiles/SimG40.yml

  3. Run the script for simulation: SimDemo.py. It will generate the simulated Bragg peak patterns, e.g. SimResult/grain_40_sim.hdf5.

• Case 2: Real Bragg peak patterns

  1. Calibrate the experimental geometry parameters, e.g. Calibration.ipynb.

  2. Write the configure file, e.g. ConfigFiles/RealG15.yml or ConfigFiles/RecG40.yml.

  3. Run the notebook: MakePeakFile.ipynb. It will generate an orientation and grain ID map, e.g. micFile/Ti7HRM2nd.hdf5, as well as a "peak file" which contains the real Bragg peak patterns of one grain, e.g. RealPeaks/RealSample_g15.hdf5.

Step 2: Reconstruction

1. Reconstruct with the script RecDemo.py or the notebook Reconstruction.ipynb.

File formats

In the whole reconstruction procedure, there are four kinds of files:

• Configure file: Files in the folder ConfigFiles/. They contain the information for simulation or reconstruction, e.g. geometry parameters, sample information, file paths, etc. Its format is described in the templates.

• Peak file: Files in the folders RealPeaks/ and SimResult/. They store the Bragg peak patterns from a single grain. They can be the output of simulation or extract from experimental images.

• Microstructure file: Files in the folder micFile/. They are the input for both reconstruction and simulation. They store the grain ID map, orientation map, and strain map (only for simulation).

• Reconstruction file: Files in the folder recFile/. They contain the reconstructed strain values.

peakFile format

It is a hdf5 file, which stores the Bragg peak patterns in windows along with other information about the experiment. As of now, the window size is fixed as (ΔJ=300, ΔK=160, ΔΩ=45), the units are number of pixels, number of pixels, and number of frames. The datasets are: (assuming there are N peaks recorded)

• "/Gs": shape of (N,3). The corresponding reciprocal vectors before distortion.

• "/MaxInt": shape of (N). The maximum intensities of peaks.

• "/OrienM": shape of (3,3). The average orientation of the grain.

• "/Pos": shape of (3). The center of mass of the grain.

• "/avg_distortion": shape of (3,3). The strain already considered in "/Gs".

• "/limits": shape of (N,5). The pixel coordinates of the window and the Ω indices of the first frame.

• "/whichOmega": shape of (N). Indicate is the first or second Bragg peak of that reciprocal vectors.

• "/Imgs/Im0": shape of (160,300,45). The diffraction pattern of the first Bragg peak.

...

An example of peak file (partial) is in the folder RealPeaks/. It is only used for demonstrating the format, so we removed the datasets "/Imgs/Im1", "/Imgs/Im2"... "/Imgs/Im91" to decrease the file size. You can also use the SimDemo.py script to generate a peak file.

micFile format

It is a hdf5 file, which contains following datasets: (assuming the mesh on sample cross section has N_x columns and N_y rows)

• "/origin": shape of (2). The position (x,y) of the bottom left corner of the mesh, in the unit of millimeter.

• "/stepSize": shape of (2). The step size (δx,δy) of the mesh.

• "/Confidence": shape of (N_y,N_x). The confidence from I9 reconstruction.

• "/Ph1": shape of (N_y,N_x). The first Euler angles on the voxels.

• "/Psi": shape of (N_y,N_x). The second Euler angles on the voxels.

• "/Ph2": shape of (N_y,N_x). The third Euler angles on the voxels.

• "/GrainID": shape of (N_y,N_x). The grain IDs on the voxels.

We also recommend store the following two datasets:

• "/Xcoordinate": shape of (N_y,N_x). The x coordinates of the voxels, in the unit of millimeter.

• "/Ycoordinate": shape of (N_y,N_x). The y coordinates of the voxels, in the unit of millimeter.

To be used for simulation, following datasets for the elastic strain components are also needed:

• "/E11": shape of (N_y,N_x).

• "/E12": shape of (N_y,N_x).

• "/E13": shape of (N_y,N_x).

• "/E22": shape of (N_y,N_x).

• "/E23": shape of (N_y,N_x).

• "/E33": shape of (N_y,N_x).

recFile format

It is a hdf5 file, which contains five important datasets: (assuming there are N voxels in the grain)

• "/Phase2_S": shape of (N,3,3). The reconstructed distortion matrix in reciprocal space on voxels. (matrix D in my thesis, p.g. 56)

• "/realS": shape of (N,3,3). The reconstructed strain tensor on voxels. (matrix V in my thesis, p.g. 55)

• "/realO": shape of (N,3,3). The reconstructed orientation matrices on voxels. (matrix R in my thesis, p.g. 55)

• "/x": shape of (N). The x coordinates of voxels, in the unit of millimeter.

• "/y": shape of (N). The y coordinates of voxels, in the unit of millimeter.

Other information about the reconstruction may also installed.

Package Structure

• Basics

  • strain_device.cu : CUDA kernel functions.
  • config.py : wrapper for configuration files.
  • initializer.py : constructs GPU related functions and objects, along with the Detector and Material objects.
  • simulator.py : simulates the Bragg peak patterns from a synthetic data, it can be used for validating the reconstruction algorithm.
  • reconstructor.py : reconstructs the intra-granular strain tensor from Bragg peak patterns.
  • optimizers.py : the minimization algorithms used in the reconstruction.

• Scripts

  • Calibration.ipynb : calibrates the geometry parameters in the experimental setup.
  • MakePeakFile.ipynb : extracts the windows around the Bragg peaks from the grain we want to reconstruct.
  • SimDemo.py : the script to simulate the Bragg peak patterns from a synthetic data, it can be used for validating the reconstruction algorithm.
  • RecDemo.py : the script to reconstruct the intra-granular strain tensor from Bragg peak patterns.
  • Reconstruction.ipynb : an alternative and more flexible way to do reconstruction instead of RecDemo.py, basically contains the procedures in reconstructor.py.

• Folders

  • AuxData/ : outputs from standard nf-HEDM, the voxelized orientations on a cross section of the sample.
  • RealPeaks/ : outputs from MakePeakFile.ipynb, the experimental Bragg peak patterns.
  • SimResult/ : outputs from SimDemo.py, the simulated Bragg peak patterns.
  • RecResult/ : the intra-granular reconstruction results.
  • micFile/ : microstructure files from FFT simulation or regenerated from the files in AuxData/.
  • ConfigFiles/ : configure files for reconstruction or simulation.
  • util/ : some basic functions related to nf-HEDM.

TODO

Because this software was a side product of a research project, many parameters are hard coded for a specific measurement of Ti7. Following parameters are currently hard coded and will be moved to the configure file in future developments:

• The ω ranges from 0 to 180 degree with 0.05 degree integration interval for each frame.
• The material has HCP crystal structure.
• The window size around the Bragg peak is fixed as (ΔJ=300, ΔK=160, ΔΩ=45).