Identification of grains with Artificial Intelligence (AI)
The GrainFind module is an important step towards precise object recognition in µCT images for accurate grain detection. With GeoDict ’s GrainFind module, individual grains can be identified in structures where the grain boundaries are previously unknown. For each identified grain, an individual best-fit shape is computed and its orientation in the structure is obtained.
In this way, simulations on the structure are possible which were impossible before, such as simulations of mechanical properties which depend on grain orientation. Furthermore, the structural information gained can be used to generate similar microstructures using the GrainGeo module, i.e. modeling the Digital Twin of the microstructure.
3D models of a material are obtained after importing and segmenting computer tomography or FIB/SEM scans of the material, and two processes are available in GrainFind:
- Identification and analysis of individual grains: grain volume, surface, surface-to-volume ratio, sphericity (shape), orientation, etc.
- Estimation of grain diameter distribution
The grain identification process finds individual grains using a watershed algorithm that was specialized to find grains by the developers at Math2Market, and a subsequent grain reconnection built up by Math2Market. For convenience, grains stemming from grains that have only been partly captured in a 3D scan, may be optionally removed on the domain boundary. These boundary grains may distort the statistics of the identified grains. Finally, a grain-shape analysis fits ellipsoids, boxes, or short fibers onto the found grains, and hence creates an analytic model for the grains.
The grain identification also provides many possibilities for post processing, such as histograms for various assessed parameters for the grains like volume, surface, and various diameters measurements.
The identified grains can be thresholded by several of the computed measurements (e.g. volume, volume-equivalent sphere diameter, inscribed-sphere diameter, sphericity, surface, or surface-to-volume ratio). This results in a grain structure with different material phases. For example, using two thresholds for the grain volume, the user is able to obtain one material for the small grains, one one material for the intermediate-sized grains, and one material for the large grains.
Furthermore, volume, surface, diameter, and various grain measures can be visualized in 3D. 3D-diameter distributions, 3D-sphericity distributions and many more results can be viewed.
The grain orientations shown in the result files are computed per voxel and stored as a 3D-orientation field. This can be used as input in ConductoDict or in ElastoDict. Furthermore, the grain identification also provides the orientation tensor, which resembles the orientation statistics and can be used in GrainGeo to reproduce such microstructures.
The grain identification statistics obtained by GrainFind and contained in the grain-identification result file (such as the diameter distribution, the orientation tensor, and others) can be directly loaded into GrainGeo's "Create Grains" with one click.
The grain-diameter estimation process computes the average grain diameter for a chosen number of different grain types, as well as its standard deviation which can be sufficient for unimodal distributions. More detailed results are provided in the form of a diameter histogram plotting the grain diameter vs. the volume fraction of grains of that diameter. Grain diameter distributions (discrete or continuous) can then be entered in the GrainGeo module to reproduce microstructure models with matching distributions.
The grain identification process in GrainFind is based on a specialization of the Watershed algorithm (widely used for the segmentation of image data) that has been developed by Math2Market. The image is converted into a distance map using the Euclidean Distance Transform (EDT). Only the parameterization of the watershed algorithm transform (choosing a minimal grain diameter) and the post-processing (reconnection of grain fragments, boundary grain handling etc.) require user input. The complexity of the algorithm – such as the EDT - is hidden “under the hood”.
- Battery materials: to identify grains in battery electrodes
- Digitalen Gesteinsphysik: to obtain information about individual grains and to gain a deeper understanding of the rock microstructure
- Filtration: to characterize filtration particles
- Verbundwerkstoffe: to detect and characterize unwanted granular particles in composite materials
- Qualitätskontrolle: for the study of heterogeneities and deviations in the shape, orientation, and size distribution of grains
- The GeoDict Base package is needed for basic functionality.
- ImportGeo-Vol: is needed to import and segment µCT images and create the microstructure models for analysis .
- GrainGeo: to model granular structures that (statistically) match the analyzed µCT images and, in this way, create Digital Twins .
- BatteryDict: Charging simulation of generated structural models of battery electrodes
- ElastoDict und ConductoDict: The grain orientation field computed by GrainFind can be used to carry out simulations of deformation and damage or thermal and electrical conductivity for transverse isotropic materials.