Multi-modal dataset of a polycrystalline metallic material: 3D microstructure and deformation fields

The development of high-fidelity mechanical property prediction models for the design of polycrystalline materials relies on large volumes of microstructural feature data. Concurrently, at these same scales, the deformation fields that develop during mechanical loading can be highly heterogeneous. Spatially correlated measurements of 3D microstructure and the ensuing deformation fields at the micro-scale would provide highly valuable insight into the relationship between microstructure and macroscopic mechanical response. They would also provide direct validation for numerical simulations that can guide and speed up the design of new materials and microstructures. However, to date, such data have been rare. Here, a one-of-a-kind, multi-modal dataset is presented that combines recent state-of-the-art experimental developments in 3D tomography and high-resolution deformation field measurements.

Methods
Material and Mechanical Testing.Wrought Inconel 718 (nominal composition in wt% Ni - 0.56%Al - 17.31%Fe - 0.14%Co - 17.97%Cr - 5.4%Nb - Ta - 1.00%Ti - 0.023%C - 0.0062%N) was subjected to a 30 minute annealing treatment at 1050 °C followed by water quenching, producing a grain size distribution centered at 62 micron with a nearly random texture. A two-step precipitation hardening treatment was conducted to form hardening precipitates. Tensile testing was performed at room temperature at a quasi-static strain rate using a custom in-situ 5000 N stage within a ThermoFisher Versa3D microscope on a flat dogbone-shaped specimen. The tensile test was interrupted at macroscopic plastic strain levels of 0.17%, 0.32%, 0.61%, and 1.26% for the collection of high-resolution images for digital image correlation (HR-DIC) measurements while loaded. 

High-Resolution Digital Image Correlation. A gold nanoparticle speckle pattern with an average particle size of 60 nanometers was deposited on the sample surface for DIC measurements. SEM image sets were acquired from the middle of the gauge length before loading and underload. Tiles of 8×8 SEM images, before and after deformation, with an image overlap of 15% were collected. Each image was acquired with a dwell time of 20 microseconds, a pixel resolution of 4096×4096, and a horizontal field width of 137 microns. Consequently, each pixel has a size of 33.4 nanometers. Regions of about 1×1 mm2 were investigated for the Inconel 718 nickel-based superalloy. DIC calculations are performed on these series of images and the results are merged using a pixel resolution merging procedure. A subset size of 31×31 pixels (1036.86×1036.86 nanometers) with a step size of 3 pixels (100.34 nanometers) was used for the DIC measurements. Digital image correlation was performed using the Heaviside-DIC method. 

3D Crystallographic Orientation Measurements. The TriBeam system is used for the collection of orientation fields in 3D over a half cubic millimeter volume. After mechanical testing , the specimen is unloaded and surface EBSD measurements are performed on the surface of the specimen on the same region where the HR-DIC measurements were made. Electrical discharge machining cuts were performed to prepare a pillar with optimal geometry for a Tribeam experiment. The pillar is laser ablated with a step size of 1 micron in Z, the sectioning direction. Between each slice, EBSD measurements are collected with a step size of 1 micron (X,Y) to form cubic voxels. A set of 526 slices was obtained during the experiment and reconstructed into a 3D dataset using the DREAM.3D software. Prior to reconstruction, each EBSD slice was aligned to match the corresponding BSE image.

Correlative measurements: Multi-modal Data Merging.The strain fields obtained from DIC corresponding to the investigated free surface of the 3D dataset are provided for the different loading steps. All fields have been aligned to fit the free surface of the 3D dataset. The distortion between both datasets was modeled using a polynomial function of degree 3. Individual slip traces were segmented from the DIC maps and indexed as individual features, using the iterative Hough transformation method. The location of each slip band in the 3D volume (coordinates of its endpoints on the (XY) surface), its inclination angle relative to the loading direction, its length, and average in-plane slip intensity and direction are all calculated. 

Mesh Generation with XtalMesh. One version of a mesh structure was created with XtalMesh, a publicly available code on GitHub. XtalMesh is used to create smooth representations of voxelized microstructures and leverages the state-of-the-art tetrahedralization algorithm fTetWild to generate an analysis-ready, boundary conforming tetrahedral mesh. The base workflow of XtalMesh was modified to better preserve the many small and thin features (mainly twins) of the Inconel 718 dataset from the effects of excessive smoothing (shrinkage and/or thinning). First, the default smoothing operation of XtalMesh was applied to the parent grain surface mesh geometry rather than of all the features/twins in the 3D dataset. This had the effect of smoothing only the twinned domains that bordered neighboring parent grains, leaving the twin boundaries still partially voxelized. At this point, the twinned regions of each parent grain are re-introduced into the parent grain mesh via constructive solid geometry (CSG) technique. For each twin, in order of smallest to largest based on a number of voxels, the intersection of its convex hull and respective parent grain mesh is computed and inserted into the overall surface mesh of the microstructure. The parent grain mesh is then redefined as the difference between itself and the previously calculated intersection. This new parent grain mesh is then used for the insertion process of the next twin. After the insertion of all twins was complete, tetrahedralization was performed on the resulting surface mesh of the entire microstructure using the fTetWild meshing algorithm. 

Geometric reconstruction and mesh generation using Simmetrix’ software suite. While it is possible to directly generate a mesh from a voxel dataset it is advantageous to introduce a geometric model, specifically a non-manifold boundary representation as an intermediate representation of the analysis domain. Such a model provides an unambiguous representation of the analysis domain and provides a mechanism to associate information such as material properties in a manner that is independent of the mesh. To be able to build a valid and appropriate (based on the needs of the simulation) finite element model from a voxel dataset assembled from a serial sectioning EBSD measurement, various procedures to remove artifacts are required. This includes the elimination of small groups of disconnected voxels, and removing noise from the grain boundaries (e.g., through the use of erosion and dilation filters). For the In718 RVE, features smaller than 50 connected voxels were removed followed by an erosion/dilation step using a 3x3x3 block structuring element. Care was taken not to apply the erosion filter to grains that were very thin (1 to 2 voxels thick) to preserve the geometry of those grains.This process is followed by the elimination of physically undesirable voxel configurations (e.g., voxel clusters of the same material connecting at a single voxel corner) that could create singularities in the finite element solution. The resulting geometric model represents each grain as a region (volume) with geometric faces (surfaces) representing grain boundaries. Attributes attached to each region allow the user to retrieve the grain ID as it was defined in the originating DREAM3D \cite{2014dream3D} dataset. At this stage, the face geometry still reflects the stair-stepped boundaries between the individual voxels, therefore a geometric-based algorithm is used to create smooth geometric faces while preserving the overall shape of the grain boundaries. The resulting geometric model can be tagged with meshing and analysis attributes to generate a run-ready input deck for the finite element solver.