Probabilistic Deep Learning Approach to Automate the Interpretation of Multi-phase Diffraction Spectra

Autonomous synthesis and characterization of inorganic materials requires the\nautomatic and accurate analysis of X-ray diffraction spectra. For this task, we\ndesigned a probabilistic deep learning algorithm to identify complex\nmulti-phase mixtures. At the core of this algorithm lies an ensemble\nconvolutional neural network trained on simulated diffraction spectra, which\nare systematically augmented with physics-informed perturbations to account for\nartifacts that can arise during experimental sample preparation and synthesis.\nLarger perturbations associated with off-stoichiometry are also captured by\nsupplementing the training set with hypothetical solid solutions. Spectra\ncontaining mixtures of materials are analyzed with a newly developed branching\nalgorithm that utilizes the probabilistic nature of the neural network to\nexplore suspected mixtures and identify the set of phases that maximize\nconfidence in the prediction. Our model is benchmarked on simulated and\nexperimentally measured diffraction spectra, showing exceptional performance\nwith accuracies exceeding those given by previously reported methods based on\nprofile matching and deep learning. We envision that the algorithm presented\nhere may be integrated in experimental workflows to facilitate the\nhigh-throughput and autonomous discovery of inorganic materials.\n