CBDR-CNN Approach for Rapid Identification of XRD Data: A Preliminary Study

Abstract

In this study, we present a novel approach combining Content-Based Data Retrieval (CBDR) and 1-dimensional Convolutional Neural Networks (1D-CNN) for crystal structure analysis of powder materials by using X-Ray Diffraction (XRD) data. The introduction sets the background by highlighting the importance of X-ray diffraction analysis and the limitations of conventional approaches in dealing with complex crystal structures. To overcome this challenge, researchers have explored artificial intelligence techniques, specifically CNN for crystal structure classification based on XRD image graph represented as 2-theta versus intensity values. The aims of this study are: implementing CBDR method on CNN model for crystal structure classification; simulating CBDR-CNN model for crystal structure classification; verifying CBDR-CNN model in crystal structure classification. Each class for CNN model such as crystal system, class material, sub-class material, and space-group achieved accuracies 99.86%, 99.99%, 99.95%, and 99.82% respectively. The results and discussion section presents the results of the CBDR-CNN model. The CBDR model effectively retrieved the most similar XRD spectrum data from the dataset based on the query properties, including Miller indices and peak position. The model effectively reduced the scope potential candidate materials, sub-materials, and space-groups. The 1D-CNN model showed high accuracy in predicting crystal properties such as material, sub-material, space-group, and crystal system. In conclusion, the CBDR-CNN approach potential revolutionizes XRD data analysis and crystal system prediction, which promotes progress in computer-aided materials study.