An AI-driven PXRD analysis framework for real-time crystal structure identification from experimental diffraction patterns.
Our system revolutionizes PXRD-based crystal identification through high-fidelity data synthesis and the cutting-edge XQueryer model. Seamlessly integrated with diffractometers, it enables precise, AI-driven material discovery and extends its capabilities to broader chemical applications. XQueryer comprises 1.03 B parameters.
Important
XQueryer is among the first AI frameworks to enable cross-system crystal structure identification directly from X-ray diffraction patterns without relying on conventional search-match pipelines.
Unlike traditional XRD analysis workflows that heavily depend on handcrafted databases, phase-by-phase retrieval, and expert-guided matching strategies, XQueryer reformulates crystal identification as a representation learning problem over diffraction space itself. By combining large-scale simulated diffraction data, deep feature alignment, and cross-system structural retrieval, XQueryer demonstrates that neural networks can learn transferable diffraction representations spanning diverse crystal systems and compositions. This work represents a major step toward universal AI-driven diffraction understanding, enabling automated crystal identification beyond fixed candidate libraries and moving XRD analysis closer to foundation-model-style structural reasoning.
-
Source Code: Available in the ./src directory.
-
Dataset: OneDrive
-
Benchmarks: Access the benchmark code at repo XqueryerBench.
-
Simulation Code: Available in the ./sim directory.
-
RRUFF–MP ID Matching: Available in the ./match directory.
-
RRUFF.db and Alex-MP-20 dataset: available in the release
- Training/Val/Testing: model_tutorial
- Simulation: sim_tutorial
- High-throughput simulation: HTsim_tutorial
Maintained by Bin Cao. Please feel free to open issues in the Github or contact Bin Cao (bcao686@connect.hkust-gz.edu.cn) in case of any problems/comments/suggestions in using the code.