2.3.3.3.111. NXem_correlation

Status:

base class, extends NXem_method

Description:

Base class to combine different method-specific data in electron microscopy. ...

Base class to combine different method-specific data in electron microscopy.

This base class represent a template for documenting correlations (spatial, temporal) between different method-specific results.

Symbols:

No symbol table

Groups cited:

NXcrystal_structure, NXdata, NXprocess

Structure:

PROCESS: (optional) NXprocess

Details about processing steps.

sequence_index: (optional) NX_INT

indexing: (optional) NXprocess

Details about correlated or logically connected EBSD datasets. ...

Details about correlated or logically connected EBSD datasets.

One important class of such correlated experiments are the so-called (quasi) in-situ experiments. In this case the same or nearly the same ROI gets analyzed via a repetitive sequence of thermomechanical treatment, sample preparation, measurement, on-the-fly-indexing. Phenomena investigated are recrystallization, strain accumulation, material damage. Post-processing is required to correlate and reidentify eventual microstructural features or local ROIs across several orientation maps.

Another important class of correlated experiments are the so-called serial-sectioning experiments. Here the same sample is measured repetitively after polishing each time, to create a stack of orientation data which can be reconstructed to a three-dimensional volume ROI.

Data can be correlated in time, position (spatial), or both (spatiotemporal).

Spatial correlations between repetitively characterized regions-of-interests are typically correlated using image registration and alignment algorithms. For this typically so-called landmarks are used. These can be grains with a very large size or specific shape, i.e. grains which are qualitatively different enough to be used as a guide how images are shifted relative to one another. Other commonly used landmarks are fiducial marks which are milled into the specimen surface using focus-ion beam milling and/or various types of indentation methods.

As far as the same physical region-of-interest is just measured several times, the additional issue of the depth increment is not a concern. However, correct assumptions for the depth increment, amount of material removed along the milling direction is relevant for accurate and precise three-dimensional (serial-sectioning) correlations. For these studies it can be tricky though to assume or estimate useful depth increments. Different strategies have been proposed like calibrations, wedged-shaped landmarks and computer simulation assisted assumption making.

Despite the use of landmarks, there are many practical issues which make the processing of correlations imprecise and inaccurate. Among these are drift and shift of the specimen, instabilities of the holder, the beam, irrespective of the source of the drift, charging effects, here specifically causing local image distortions and rotations which may require special processing algorithms to reduce such imprecisions.

Time correlations face all of the above-mentioned issues surplus the challenge that specific experimental protocols have to be used to ensure the material state is observed at specific physical time. The example of quasi in-situ characterization of crystal growth phenomena, a common topic in engineering or modern catalysis research makes it necessary to consider that e.g. the target value for the desired annealing temperature is not just gauged based on macroscopic arguments but considers that transient effects take place. Heating or quenching a sample might thus might not have been executed under conditions in the interaction volume as they are documented and/or assumed.

These issue cause that correlations have an error margin as to how accurately respective datasets were not only just synced based on the geometry of the region-of-interests and the time markers but also to asssure which physical conditions the specimen experienced over the course of the measurements.

The fourth example of the em_om reference implementation explores the use of the correlation group with a serial-sectioning datasets that was collected by the classical Inconel 100 dataset collected by M. D. Uchic and colleagues (M. Groeber M, Haley BK, Uchic MD, Dimiduk DM, Ghosh S 3d reconstruction and characterization of polycrystalline microstructures using a fib-sem system data set. Mater Charac 2006, 57 259–273. 10.1016/j.matchar.2006.01.019M).

This dataset was specifically relevant in driving forward the implementation of the DREAM.3D software. DREAM.3D is an open-source software project for post-processing and reconstructing, i.e. correlating sets of orientation microscopy data foremost spatially. One focus of the software is the (post-)processing of EBSD datasets. Another cutting edge tool with similar scope but a commercial solution by Bruker is QUBE which was developed by P. Konijnenberg and coworkers.

Conceptually, software like DREAM.3D supports users with creating linear workflows of post-processing tasks. Workflows can be instructed via the graphical user interface or via so-called pipeline processing via command line calls. DREAM.3D is especially useful because its internal system documents all input, output, and parameter of the processing steps. This makes DREAM.3D a good candidate to interface with tools like em_om parser. Specifically, DREAM.3D documents numerical results via a customized HDF5 file format called DREAM3D. Workflow steps and settings are stored as nested dictionaries in JSON syntax inside a supplementary JSON file or alongside the data in the DREAM3D file. DREAM.3D has a few hundred algorithms implemented. These are called filters in DREAM.3D terminology.

Users configure a workflow which instructs DREAM.3D to send the data through a chain of predefined and configured filters. Given that for each analysis the filter is documented via its version tags surplus its parameter and setting via a controlled vocabulary, interpreting the content of a DREAM3D HDF5 file is possible in an automated manner using a parser. This makes DREAM.3D analyses repeatable and self-descriptive. A key limitation though is that most frequently the initial set of input data come from commercial files like ANG. This missing link between the provenance of these input files, their associated creation as electron microscope session, is also what NXem_ebsd solves.

Nevertheless, as this can be solved with e.g. NXem_ebsd we are convinced that the DREAM.3D and the em_om parser can work productively together to realize RDMS-agnostic parsing of serial-section analyses.

The internal documentation of the DREAM.3D workflow also simplifies the provenance tracking represented by an instance of NXem_ebsd as not every intermediate results has to be stored. Therefore, the fourth example focuses on the key result obtained from DREAM.3D - the reconstructed and aligned three-dimensional orientation map.

Usually, this result is the starting point for further post-processing and characterization of structural features. As here orientation microscopy is insofar scale invariant using DREAM.3D, NXem_ebsd, and em_om should be useful for different characterization methods, such as EBSD, Transmission Kikuchi Diffraction (TKD), Automated Crystal Orientation Mapping (ACOM), Nanobeam Electron Diffraction (using commercial systems like NanoMegas ASTAR) or open-source implementations of these techniques (such as via pyxem/orix).

The result of orientation microscopy methods are maps of local orientation and thermodynamic phase (crystal structure) pieces of information. Virtually all post-processing of such results for structural features includes again a workflow of steps which are covered though by the NXmicrostructure partner application definition. The respective source of the data in an instance of NXmicrostructure can again be a link or reference to an instance of NXem_ebsd to complete the chain of provenance.

CRYSTAL_STRUCTURE: (optional) NXcrystal_structure

roi: (optional) NXdata

descriptor: (optional) NX_CHAR

Descriptor representing the image contrast.

title: (optional) NX_CHAR

Title of the default plot.

data: (optional) NX_NUMBER (Rank: 3, Dimensions: [n_z, n_y, n_x]) {units=NX_UNITLESS}

Descriptor values displaying the ROI.

@long_name: (optional) NX_CHAR

Descriptor values.

axis_z: (optional) NX_NUMBER (Rank: 1, Dimensions: [n_z]) {units=NX_LENGTH}

Calibrated coordinate along the z-axis.

@long_name: (optional) NX_CHAR

Label for the z axis

axis_y: (optional) NX_NUMBER (Rank: 1, Dimensions: [n_y]) {units=NX_LENGTH}

Calibrated coordinate along the y-axis.

@long_name: (optional) NX_CHAR

Label for the y axis

axis_x: (optional) NX_NUMBER (Rank: 1, Dimensions: [n_x]) {units=NX_LENGTH}

Calibrated coordinate along the x-axis.

@long_name: (optional) NX_CHAR

Label for the x axis

Hypertext Anchors

List of hypertext anchors for all groups, fields, attributes, and links defined in this class.

NXDL Source:

https://github.com/FAIRmat-NFDI/nexus_definitions/tree/fairmat/contributed_definitions/NXem_correlation.nxdl.xml