Abstract
The design of novel materials hinges on the understanding ofstructure-property relationships. However, in recent times, our capability tosynthesize a large number of materials has outpaced our speed at characterizingthem. While the overall chemical constituents can be readily known duringsynthesis, the structural evolution and characterization of newly synthesizedsamples remains a bottleneck for the ultimate goal of high throughputnanomaterials discovery. Thus, scalable methods for crystal symmetrydetermination that can analyze a large volume of material samples within ashort time-frame are especially needed. Kikuchi diffraction in the SEM is apromising technique for this due to its sensitivity to dynamical scattering,which may provide information beyond just the seven crystal systems andfourteen Bravais lattices. After diffraction patterns are collected frommaterial samples, deep learning methods may be able to classify the space groupsymmetries using the patterns as input, which paired with the elementalcomposition, would help enable the determination of the crystal structure. Toinvestigate the feasibility of this solution, neural networks were trained topredict the space group type of background corrected EBSD patterns. Ournetworks were first trained and tested on an artificial dataset of EBSDpatterns of 5,148 different cubic phases, created through physics-baseddynamical simulations. Next, Maximum Classifier Discrepancy, an unsuperviseddeep learning-based domain adaptation method, was utilized to train neuralnetworks to make predictions for experimental EBSD patterns. We introduce arelabeling scheme, which enables our models to achieve accuracy scores higherthan 90% on simulated and experimental data, suggesting that neural networksare capable of making predictions of crystal symmetry from an EBSD pattern.