Using kikuchipy for phase identification of zirconium

[16sar]

Hi,

My team's project involves using EBSD to identify α-Zr (HCP) and β-Zr (BCC) phases in our samples. We are using Bruker's ESPRIT software for indexing the kikuchi diffraction patterns and it is having trouble identifying the β-Zr phase, so we want to try other methods. I noticed that HCP is not an option for the phaselist parameter in index_pats() - is there any way to use kikuchipy to index HCP diffraction patterns, or is this not within the scope of the library? If not, do you have any suggestions for other libraries that can identify HCP patterns?

Thank you

[hakonanes]

Hi @16sar,

thank you for considering kikuchipy and PyEBSDIndex to solve your issue here. Note that kikuchipy depends on PyEBSDIndex for Hough indexing and the syntax you refer to comes from the latter.

I noticed that HCP is not an option for the phaselist parameter in index_pats()

You're right, the released version of PyEBSDIndex has interplanar angle look-up tables only for FCC and BCC. I know @drowenhorst-nrl has been working to support indexing of patterns from crystals of the remaining nine Laue classes, but I don't know how close these are to a release.

is there any way to use kikuchipy to index HCP diffraction patterns

Yes, but then you need (dynamical) EBSD master pattern simulations (full Kikuchi sphere) of alpha and beta Zr, produced with e.g. EMsoft (free) or Bruker DynamicS (license). You could then adapt the pattern matching tutorial to your problem for indexing.

or is this not within the scope of the library?

No, indexing all EBSD patterns is within the scope of both PyEBSDIndex and kikuchipy. Things take time... that's why it's not supported yet.

If not, do you have any suggestions for other libraries that can identify HCP patterns?

Much of the pattern matching capabilities in kikuchipy relies on the work of the creators of EMsoft; you could try that package for dictionary indexing followed by refinement (what kikuchipy also supports). EMsoft is a suite of Fortran programs, installable on any OS (pre-built binaries for Windows and macOS, some more steps are needed to install on Linux).

If you have a MATLAB license, you could try AstroEBSD. They support both Hough indexing and so-called refined template matching. The latter requires EBSD master pattern simulations. Their Hough indexing should support other Laue classes than the cubic ones, but I haven't tested it myself.

[drowenhorst-nrl]

@hakonanes covered the questions quite well, and I will add just a bit more info. The "develop" branch of PyEBSDIndex does have support for the other crystal classes, and at least my initial tests for HCP appear to perform well. The only significant note I will add is that I am pretty sure there is a conflict in how the crystal orientation is defined for some vendors. PyEBSDIndex uses the EDAX/EMsoft/kikuchipy [edited to put kikuchipy in this list] convention of aligning the a_1 axis of the crystal with the x-axis of the crystal reference frame, while (I think) Oxford?/Bruker all use the convention of aligning the c-axis with the z-axis. The result is that for hexagonal, there is a 30° mismatch in the orientations. I need to do some thinking/testing to make sure I have those differences properly corrected (not just for hex).

And just want to define expectations ... I have very little exposure to the Bruker systems, and thus cannot give a good answer if PyEBSDIndex actually performs better than they do in indexing. I can say that in terms of successful indexing with poor patterns, PyEBSDIndex is very comparable to EDAX (I will note that the accuracy and precision of PyEBSDIndex is considerably improved over EDAX, but indexing success rate is similar). EMsoft Dictionary indexing is significantly more robust to low quality patterns.

Alternatively, you may want to attempt to improve pattern quality using NLPAR - a pattern denoising algorithm that is also included in the PyEBSDIndex package. We are still doing some testing there, but we are finding that no matter the indexing method, NLPAR will improve the results (with the note that it is not magic, there has to be something in the patterns to improve).

[hakonanes]

Thank you for pitching in, @drowenhorst-nrl.

kikuchipy uses orix for handling of the crystal symmetry of a phase, which uses the same crystal axes alignment as EMsoft. That is the intention, at least... Could you point me to what makes you think kikuchipy uses another alignment?

[drowenhorst-nrl]

I sit corrected! kikuchipy/orix uses the same convention that (I was pretty sure) we are using. Sorry for the confusion. @hakonanes maybe we need to chat off-line to make sure we have everything in alignment (I thought we had a conflict on hexagonal orientations in some testing, but maybe that is arising from somewhere else?)

[hakonanes]

Well, I think you might be right that the hexagonal alignment is not exactly as you use in PyEBSDIndex (EDAX/EMsoft). There are a few outstanding issues related to crystal symmetry in orix yet which we haven't sorted out. When I get to that (next month), I'll ask you if you're available, thanks.

[16sar]

@hakonanes @drowenhorst-nrl Thank you both for your responses. I'll see if I can use our Bruker software to generate the EBSD master patterns so I can try the pattern matching method, and I will also try using NPLAR to process the patterns.

[hakonanes]

Sounds good.

Pattern matching in kikuchipy requires that the master pattern is an instance of the EBSDMasterPattern class in the square-Lambert projection (see docs). As Dave, I'm not that familiar with Bruker's file formats (we can read patterns from their HDF5 files, and hopefully from the .bcf format in the future). I believe it's possible to export master patterns from DynamicS as image files. The procedure would then be:

1. Load image with HyperSpy's hs.load("image_file.png", signal_type="EBSDMasterPattern", phase=phase, hemisphere="both", projection="stereographic"). phase is here a Phase instance from orix, which you have to load from a cif via Phase.from_cif() or create manually. I don't know Zr so well. If your phases are centrosymmetric, you just need the upper hemisphere simulation (hemisphere="upper", the default). If not, you need to pass both hemispheres when creating the EBSDMasterPattern instance. Finally, I can't remember if DynamicS allows exporting simulations in the square-Lambert projection. If yes, I'd do that and then you can load them directly (projection="lambert"). If not, load it in the stereographic projection but convert to Lambert projection in a subsequent step.
2. Potentially convert from stereographic projection to square-Lambert projection with EBSDMasterPattern.as_lambert().

If I remember correctly, DynamicS allows exporting master patterns in a binary .bin file. We cannot read this in kikuchipy, but if you send such a file to me with all the metadata (how it was made) and the expected format of the data, I'll try to quickly make a reader for it so that the above can be done in one or two lines of code instead (assuming that phase information is stored in the binary file).

Please report back here if this procedure doesn't work for you, or you need any guidance.

[hakonanes]

@16sar, just-released PyEBSDIndex v0.2.0 now supports indexing patterns from HCP crystals. The updated tutorial explains how to describe the phases for indexing (not just "FCC" or "BCC" anymore): https://pyebsdindex.readthedocs.io/en/stable/tutorials/ebsd_index_demo.html.

Note that kikuchipy is not compatible with this version yet. We're working on this in #652.

[hakonanes]

@16sar, and now just-released kikuchipy v0.9.0 supports Hough indexing of all Laue classes (also 6/mmm) via PyEBSDIndex. I recommend you create your own reflector list, though. I've started a discussion here. I also recommend you check out our kinematical EBSD simulation tutorial for more info on creating reflector lists.