Deep Neural Network Enabled Space Group Identification in EBSD

Abstract

Electron backscatter diffraction (EBSD) is one of the primary tools in materials development and analysis. The technique can perform simultaneous analyses at multiple length scales, providing local sub-micron information mapped globally to centimeter scale. Recently, a series of technological revolutions simultaneously increased diffraction pattern quality and collection rate. After collection, current EBSD pattern indexing techniques (whether Hough-based or dictionary pattern matching based) are capable of reliably differentiating between a “user selected” set of phases, if those phases contain sufficiently different crystal structures. EBSD is currently less well suited for the problem of phase identification where the phases in the sample are unknown. A pattern analysis technique capable of phase identification, utilizing the information-rich diffraction patterns potentially coupled with other data, such as EDS-derived chemistry, would enable EBSD to become a high-throughput technique replacing many slower (X-ray diffraction) or more expensive (neutron diffraction) methods. We utilize a machine learning technique to develop a general methodology for the space group classification of diffraction patterns; this is demonstrated within the $\lpar 4/m\comma \;\bar{3}\comma \;\;2/m\rpar$ point group. We evaluate the machine learning algorithm's performance in real-world situations using materials outside the training set, simultaneously elucidating the role of atomic scattering factors, orientation, and pattern quality on classification accuracy.

An Acquisition Parameter Study for Machine-Learning-Enabled Electron Backscatter Diffraction

“…However, the CNN's classification accuracy is observed to be quite stable (i.e., small reductions in classification accuracy) in comparison to the results achieved with the default conditions, suggesting that the features detectors learned by the model are not biased to these characteristics. This was one of the intended goals of using multiple materials with different z-contrast and lattice parameters for the same space group in the training set (Kaufmann et al, 2020d), and this study indicates its effectiveness. Moreover, the CNN is also observed to be highly dependable after decreasing the signal-to-noise ratio of the captured diffraction pattern by reducing the frame averaging.…”

“…In each of the studies within this work, the space groups most likely to be misclassified by the model were 221 (Pm 3m ) and 229 (Im 3m ). As previously mentioned, the misclassification of these patterns can be at least partially attributed to the strong similarities between diffraction patterns from the fcc (Fm 3m) and L1 2 (Pm 3m) structures and bcc (Im 3m) and B2 (Pm 3m) structures used in training the CNN (Kaufmann et al, 2020d). Inclusion of more diverse data for these space groups may help alleviate this concern in addition to being a practical advancement toward commercial adoption.…”

“…Refer to Figure 5 in Xception: Deep Learning with Depthwise Separable Convolutions (Chollet, 2017) for a complete description of the Xception architecture. Further details about the training process can be found in Kaufmann et al (2020d). The CNNs were implemented with TensorFlow (Abadi et al, 2016) and Keras (Chollet, 2015).…”

“…The fields of materials development and analysis have recently begun to explore the possibility of applying data-driven strategies and artificial intelligence (AI) for accelerating or automating a variety of tasks (Ong et al, 2013;O'Mara et al, 2016;Dagdelen et al, 2017;Ward et al, 2018;Ziletti et al, 2018;Avery et al, 2019;DeCost et al, 2019;Oviedo et al, 2019;Tshitoyan et al, 2019;Dunn et al, 2020;Kaufmann & Vecchio, 2020;Kaufmann et al, 2020aKaufmann et al, , 2020dStan et al, 2020;Chen et al, 2021). Computer vision is a subset of AI with the goal of training computers to understand the visual world and potentially act on that information (Zhu et al, 2020a(Zhu et al, , 2020b.…”

Autonomous EBSD Pattern Classification Performance with Changing Acquisition Parameters

“…In material analysis, these tools have largely been applied to techniques requiring analysis of data collected in the form of images [1,2]. Electron backscatter diffraction (EBSD) is one such technique benefitting from these recent efforts to improve material analysis by leveraging deep neural networks [3][4][5][6][7]. EBSD is an SEM-based technique involving the capture of 2D diffraction patterns from the surface of a well-polished sample [8].…”