Quantification of microstructures is crucial for understanding processing–structure and structure–property relationships in polycrystalline materials. Delineating grain boundaries in bright-field transmission electron micrographs, however, is challenging due to complex diffraction contrast in images. Conventional edge detection algorithms are inadequate; instead, manual tracing is usually required. This study demonstrates the first successful machine learning approach for grain boundary detection in bright-field transmission electron micrographs. The proposed methodology uses a U-Net convolutional neural network trained on carefully constructed data from bright-field images and hand tracings available from prior studies, combined with targeted postprocessing algorithms to preserve fine features of interest. The image processing pipeline accurately estimates grain boundary positions, avoiding segmentation in regions with intragrain contrast and identifying low-contrast boundaries. Our approach is validated by directly comparing microstructural markers (i.e., grain centroids) identified in U-Net predictions with those identified in hand tracings; furthermore, the grain size distributions obtained from the two techniques show notable overlap when compared using t-test, Kolmogorov–Smirnov test, and Cramér–von Mises test. The technique is then successfully applied to interpret new microstructures having different image characteristics from the training data, with preliminary results from platinum and palladium microstructures presented, highlighting the versatility of our approach for grain boundary identification in bright-field micrographs.
Metallic thin films offer a platform to experimentally study the dynamics of microstructural evolution, but the required transmission electron microscopy (TEM)-based imaging generates complex images that are challenging to segment and quantify. This work provides a comparative analysis of a new YOLOv8 model and an established U-Net model for bright-field TEM images of polycrystals, employing a framework leveraging physical observables to evaluate performance against two hand-traced benchmark datasets. This methodology obviates the comparison of large, diversely structured, and manually labeled datasets that are required to assess performance on a per-image/per-pixel basis. It is found that the YOLOv8 model, adapted for real-time instance segmentation, has up to 43× faster inferencing (NVIDIA GeForce RTX 4090) compared to U-Net and reconstructs hand-traced grain size distributions (GSDs) with excellent fidelity, finding mean diameter within 3% for grains near an optimal magnification; for grains that deviate from the optimal pixel-diameter, the size of small- (large)-diameter grains is systematically over- (under)-estimated. This is partially mitigated by including scale-aware augmentations during training. Moreover, when the bias is corrected post-inference by a rigid shift in distribution, the YOLOv8 model reproduces ground truth GSDs with exceptional fidelity, with statistical tests indicating <5% probability that the distributions are distinct. Based on ground truth data, calibration curves pertaining to this shift can be constructed for a given model. This issue is not present in the U-Net model’s results, indicating that for quantitative measurements where the true size of objects is of interest, special procedures must be implemented for YOLO-based models.
This work outlines conditions suitable for the heteroepitaxial growth of Cr2O3(0001) films (1.5–20 nm thick) on a Ru(0001)-terminated substrate. Optimized growth is achieved by sputter deposition of Cr within a 4 mTorr Ar/O2 20% ambient at Ru temperatures ranging from 450 to 600 °C. The Cr2O3 film adopts a 30° rotated honeycomb configuration with respect to the underlying Ru(0001) substrate and exhibits a hexagonal lattice parameter consistent with that for bulk Cr2O3(0001). Heating to 700 °C within the same environment during film preparation leads to Ru oxidation. Exposure to temperatures at or above 400 °C in a vacuum, Ar, or Ar/H2 3% leads to chromia film degradation characterized by increased Ru 3d XPS intensity coupled with concomitant Cr 2p and O 1s peak attenuations when compared to data collected from unannealed films. An ill-defined but hexagonally well-ordered RuxCryOz surface structure is noted after heating the film in this manner. Heating within a wet Ar/H2 3% environment preserves the Cr2O3(0001)/Ru(0001) heterolayer structure to temperatures of at least 950 °C. Heating an Ru–Cr2O3–Ru heterostacked film to 950 °C within this environment is shown by cross-sectional scanning/transmission electron microscopy (S/TEM) to provide clear evidence of retained epitaxial bicrystalline oxide interlayer structure, interlayer immiscibility, and epitaxial registry between the top and bottom Ru layers. Subtle effects marked by O enrichment and O 1s and Cr 2p shifts to increased binding energies are noted by XPS in the near-Ru regions of Cr2O3(0001)/Ru(0001) and Ru(0001)/Cr2O3(0001)/Ru(0001) films after annealing to different temperatures in different sets of environmental conditions.
Machine learning strategies for the semantic segmentation of materials’ micrographs, such as U-Net, have been employed in recent years to enable the automated identification of grain-boundary networks in polycrystals. For example, most recently, this architecture has allowed researchers to address the long-standing problem of automated image segmentation of thin-film microstructures in bright-field TEM micrographs. Such approaches are typically based on the minimization of a binary cross-entropy loss function that compares constructed images to a ground truth at the pixel level over many epochs. In this work, we quantify the rate at which the underlying microstructural features embodied in the grain-boundary network, as described stereologically, are also learned in this process. In particular, we assess the rate of microstructural learning in terms of the moments of the k-th nearest-neighbor pixel distributions and associated metrics, including a microstructural cross-entropy, that embody the spatial correlations among the pixels through a hierarchy of n-point correlation functions. From the moments of these distributions, we obtain so-called learning functions that highlight the rate at which the important topological features of a grain-boundary network appear. It is found that the salient features of network structure emerge after relatively few epochs, suggesting that grain size, network topology, etc., are learned early (as measured in epochs) during the segmentation process.