Integrated edge-to-exascale workflow for real-time steering in neutron scattering experiments

Area of Science:

• Materials Science
• Computational Physics
• Data Science

Background:

• Neutron scattering experiments generate vast amounts of complex data.
• Real-time analysis and adaptive control are crucial for optimizing experimental efficiency and accuracy.
• Current data processing methods can be time-consuming, limiting experimental throughput.

Purpose of the Study:

• To introduce a novel computational framework for real-time steering of neutron scattering experiments.
• To leverage artificial intelligence (AI), machine learning (ML), and high-performance computing (HPC) for enhanced experimental control.
• To reduce data processing bottlenecks and improve the accuracy of multidimensional crystallography.

Main Methods:

• Integration of edge and exascale computing for an edge-to-exascale workflow.
• Application of the Temporal Fusion Transformer model for predicting 3D neutron scattering patterns with voxel-level precision.
• Temporal processing of four-dimensional neutron event data from the Spallation Neutron Source.

Main Results:

• Demonstrated substantial reduction in data processing time from hours to minutes using distributed training.
• Achieved significant improvements in model accuracy for predicting neutron scattering patterns.
• Enabled adaptive, data-driven decisions during experiments through edge and exascale computing.

Conclusions:

• The developed framework optimizes neutron beam time and improves experimental accuracy.
• This approach lays the foundation for automation in neutron scattering experiments.
• The system shows significant potential for widespread adoption and efficient exploration of complex material systems.