View in EDS

GCN-Based Framework for Materials Screening and Phase Identification

Saved in:
Bibliographic Details
Title: GCN-Based Framework for Materials Screening and Phase Identification
Authors: Zhenkai Qin, Qining Luo, Weiqi Qin, Xiaolong Chen, Hongfeng Zhang, Cora Un In Wong
Source: Materials ; Volume 18 ; Issue 5 ; Pages: 959
Publisher Information: Multidisciplinary Digital Publishing Institute
Publication Year: 2025
Collection: MDPI Open Access Publishing
Subject Terms: X-ray diffraction pattern analysis, graph-based phase identification, deep learning for crystallography, diffraction peak correlation
Description: This study proposes a novel framework using graph convolutional networks to analyze and interpret X-ray diffraction patterns, addressing challenges in phase identification for multi-phase materials. By representing X-ray diffraction patterns as graphs, the framework captures both local and global relationships between diffraction peaks, enabling accurate phase identification even in the presence of overlapping peaks and noisy data. The framework outperforms traditional machine learning models, achieving a precision of 0.990 and a recall of 0.872. This performance is attained with minimal hyperparameter tuning, making it scalable for large-scale material discovery applications. Data augmentation, including synthetic data generation and noise injection, enhances the model’s robustness by simulating real-world experimental variations. However, the model’s reliance on synthetic data and the computational cost of graph construction and inference remain limitations. Future work will focus on integrating real experimental data, optimizing computational efficiency, and exploring lightweight architectures to improve scalability for high-throughput applications.
Document Type: text
File Description: application/pdf
Language: English
Relation: Materials Simulation and Design; https://dx.doi.org/10.3390/ma18050959
DOI: 10.3390/ma18050959
Availability: https://doi.org/10.3390/ma18050959
Rights: https://creativecommons.org/licenses/by/4.0/
Accession Number: edsbas.1D3F1277
Database: BASE
Description
Abstract:This study proposes a novel framework using graph convolutional networks to analyze and interpret X-ray diffraction patterns, addressing challenges in phase identification for multi-phase materials. By representing X-ray diffraction patterns as graphs, the framework captures both local and global relationships between diffraction peaks, enabling accurate phase identification even in the presence of overlapping peaks and noisy data. The framework outperforms traditional machine learning models, achieving a precision of 0.990 and a recall of 0.872. This performance is attained with minimal hyperparameter tuning, making it scalable for large-scale material discovery applications. Data augmentation, including synthetic data generation and noise injection, enhances the model’s robustness by simulating real-world experimental variations. However, the model’s reliance on synthetic data and the computational cost of graph construction and inference remain limitations. Future work will focus on integrating real experimental data, optimizing computational efficiency, and exploring lightweight architectures to improve scalability for high-throughput applications.
DOI:10.3390/ma18050959