Powder X‐Ray Diffraction Pattern Is All You Need for Machine‐Learning‐Based Symmetry Identification and Property Prediction

Abstract

Herein, data‐driven symmetry identification, property prediction, and low‐dimensional embedding from powder X‐Ray diffraction (XRD) patterns of inorganic crystal structure database (ICSD) and materials project (MP) entries are reported. For this purpose, a fully convolutional neural network (FCN), transformer encoder (T‐encoder), and variational autoencoder (VAE) are used. The results are compared to those obtained from a well‐established crystal graph convolutional neural network (CGCNN). A task‐specified small dataset that focuses on a narrow material system, knowledge (rule)‐based descriptor extraction, and significant data dimension reduction are not the main focus of this study. Conventional powder XRD patterns, which are most widely used in materials research, can be used as a significantly informative material descriptor for deep learning. Both the FCN and T‐encoder outperform the CGCNN for symmetry classification. For property prediction, the performance of the FCN concatenated with multilayer perceptron reaches the performance level of CGCNN. Machine‐learning‐driven material property prediction from the powder XRD pattern deserves appreciation because no such attempts have been made despite common XRD‐driven symmetry (and lattice size) prediction and phase identification. The ICSD and MP data are embedded in the 2D (or 3D) latent space through the VAE, and well‐separated clustering according to the symmetry and property is observed.

Machine Learning Automated Analysis of Enormous Synchrotron X-ray Diffraction Datasets

“…The poor performance of the models trained without labeled data can be attributed to the significant differences between the synthetic XRD patterns used in training and the μ-XRD data obtained from experiments. In previous studies, experimental data used for evaluating DNN models were typically collected from solid samples in ambient environments. − In such cases, high-quality X-ray diffraction peaks from different planes were easily obtainable, which could be fitted to the synthetic data generated from the crystallographic structure. However, μ-XRD data obtained from hydrothermal fluid systems only exhibits a few primary diffraction peaks with low-index ( hkl ) lattice planes, even under the best circumstances, due to several adverse factors, including the small diffraction volume with microbeam, low exposure time, poorly crystalline samples, preferred orientation or large crystal compared with beam size, and overexposure in 2D images (extra bright spot, see Figure S6).…”

“…The theoretical XRD patterns were generated by mixing two and three end member patterns (bastnaesite, calcite, and rhenium (Re) metal). This mixing was conducted multiple times for every combination of phases with different ratios. − Small number of labeled experimental XRD patterns were also produced for generating data sets (see additional information in the Supporting Information for detailed composition of training data sets).…”

“…All DNN models trained in this study used multiple layers of convolution and maxpooling to extract the features from the 1D XRD pattern. Based on the previously studies, the combination of these two layers improves the accuracy of models. − As shown in Figure S1, the binary classification DNN models determine the existence of phases from input XRD pattern mainly consists of 3 different kinds of layers including a 1d convolution layer, a maxpooling layer, and a fully connected layer. At the start of the model, normalized data points are fed into a 1d convolution layer (input channel of 1, output channel of 4, kernel size of 5, stride size of 1, and padding size of 1) which effectively extracts the features from the input.…”

“…A promising solution to the challenges posed by massive XRD data sets is the use of deep neural network (DNN) based models, which can automatically extract information from XRD patterns. − These models leverage the advantages of combining 1D convolution layers and dense layers to extract and learn features from labeled XRD patterns in the training data set. Several studies have reported using DNN models to extract various types of information from experimental XRD patterns, such as phase identity and ratio, unit cell parameters, and even the ability to distinguish potential perovskite materials. − However, since DNN model training usually requires a large amount of diverse data, a common practice is to generate theoretical XRD patterns from crystallographic structures. − Previous studies have shown that models trained on theoretical XRD patterns can analyze XRD patterns collected from solid samples in ambient environments. − However, it remains unclear whether these models can also accurately and efficiently analyze data collected from the μ-XRD mapping of in situ experiments involving complex mixtures of solids and liquid.…”

DeepXRD, a Deep Learning Model for Predicting XRD spectrum from Material Composition

“…In this paper, we aim to explore whether we can develop a deep learning algorithm to predict the XRD spectrum from the composition alone, which can then be used for fast large-scale structure-oriented screening in modern computational generative materials design. Since our predicted XRDs can be fed to downstream algorithms to predict their structural dimensionality, crystal systems, and space groups, our XRD prediction algorithm can be very useful for screening potential new materials with only their composition information, which can significantly narrow down the crystal structure prediction and Density Functional Theory (DFT) calculations effort. Our XRD prediction algorithm can also be potentially used for the unsupervised discovery of new materials with similar properties to known materials .…”