Session: 16-01-01: Government Agency Student Poster Competition

Paper Number: 151033

151033 - Advancing Automated Classification of Crystallographic Structures Using Synthetic Two-Dimensional X-Ray Diffraction Patterns and Deep Learning 

Two-Dimensional X-Ray Diffraction (2D XRD) represents a significant advancement in the analysis of material structures, providing more comprehensive data compared to traditional one-dimensional methods. The primary aim of this research is to develop and utilize synthetic 2D XRD spot patterns for the automated classification of crystal systems, space groups, and point groups using advanced deep learning techniques. This approach is motivated by the need to overcome the limitations posed by the scarcity of large experimental 2D datasets, which are critical for training robust machine learning models.

Our work contributes to the field of materials science and engineering by creating an extensive synthetic dataset derived from 177,000 CIF entries in the Inorganic Crystal Structure Database (ICSD). This dataset serves as the foundation for developing and testing our deep learning models. The core of our methodology is the AutoDiffraction Pipeline (ADP), a novel system that transforms real space data into reciprocal space, defines zonal regions, and calculates diffraction intensities. This pipeline is designed to generate realistic 2D XRD patterns that mimic experimental data, thereby providing a rich source of information for training deep learning models. In addition, These zonal regions are critical as they represent specific orientations of the crystal lattice that are relevant for diffraction experiments.

Once the synthetic 2D XRD patterns are generated, they are fed into a deep learning framework designed for classification tasks. This framework includes convolutional neural networks (CNNs) that are particularly well-suited for image recognition and classification. The deep learning models are trained on the synthetic data to learn the intricate patterns and features associated with different crystal systems, space groups, and ponit groups.

Preliminary results from our experiments indicate that the deep learning models, trained on our synthetic dataset, can achieve high accuracy in classifying. This success demonstrates the effectiveness of our approach in addressing the data scarcity issue and highlights the potential of synthetic data in enhancing machine learning applications in materials science.

In conclusion, this research presents a significant step forward in the automated classification of crystal systems, space groups, and point groups. By generating a comprehensive synthetic dataset and developing a robust deep learning framework, we provide a valuable tool for the materials science community. This work not only improves the efficiency and accuracy of materials analysis but also opens new avenues for the application of artificial intelligence in scientific research. Future work will focus on refining the ADP, expanding the synthetic dataset, and further improving the deep learning models to enhance their generalizability and performance on experimental data.

Presenting Author: Ayoub Shahnazari University of Rochester

Presenting Author Biography: My name is Ayoub Shahnazari. I am currently a PhD student in Mechanical Engineering at the University of Rochester. I have received my bachelor’s and master’s degrees in Mechanical Engineering from the University of Guilan, Rasht, Iran. My research theme can be divided into XRD (X-ray Diffraction), FEM (Finite Element Method), CAD (computer-aided design), and DFT (Density Functional Theory). In my master’s thesis, I investigated the elastic and plastic properties of various nanostructures using DFT and FEM under different boundary conditions.

Now, as a PhD student at the University of Rochester, my focus has expanded to include deep learning and its applications in analyzing X-ray diffraction patterns. My current research involves developing and utilizing synthetic 2D XRD patterns for the automated classification of crystal systems, space groups, and point groups using advanced deep learning techniques.

Authors:

Ayoub Shahnazari University of Rochester
Zeliang Zhang University of Rochester
Sachith Dissanayake University of Rochester
Chenliang Xu University of Rochester
Niaz Abdolrahim University of Rochester