Niaz Abdolrahim1,Jerardo Salgado1,Zhaotong Du1,Samuel Lerman1,Ayoub Shahnazari1,Zeliang Zhang1,Chenliang Xu1
University of Rochester1
Niaz Abdolrahim1,Jerardo Salgado1,Zhaotong Du1,Samuel Lerman1,Ayoub Shahnazari1,Zeliang Zhang1,Chenliang Xu1
University of Rochester1
The development of new technologies such as superconductors, fusion energy, and superfast quantum computers heavily relies on a deep understanding of material properties. X-ray diffraction (XRD) is a powerful technique used for material characterization, providing valuable information about the atomic symmetry of materials. However, despite recent advancements in dynamic in-situ XRD experiments, fully characterizing these materials remains challenging due to the large amounts of data generated. This data contains crucial insights into material structure, behavior, and transformation pathways that could otherwise be overlooked.<br/>To address this challenge, we have developed deep learning models integrated with physics-based reasoning, enabling automatic classification of the crystal system and space group of 1D XRD data. Our models were trained using a comprehensive dataset consisting of hundreds of thousands of diffraction images. These images were generated using our Python-based pipeline, which incorporates fundamental principles such as Bragg's Law and atomic scattering factors.<br/>By leveraging high-quality training and evaluation data, we achieved accurate classification based on Bragg's Law and improved performance when applied to experimental data. Our models provide instantaneous feedback and can classify data without requiring human intervention, making them highly efficient for extracting valuable material insights from large XRD datasets.<br/>Moreover, the versatility of our models extends beyond XRD data. They can be applied to other spectral data, such as Raman spectroscopy or nuclear magnetic resonance spectroscopy, broadening their applicability to diverse material characterization techniques.<br/>By combining the power of deep learning with physics-based justification, our approach enables efficient and accurate analysis of material properties, paving the way for accelerated advancements in various technological fields that rely on a deep understanding of materials.