Efficiency enhancement by autonomous “closed-loop” synchrotron and neutron scattering experiments enabled by machine learning

• Applicant: Professor Dr. Frank Schreiber
• Subject Area: Experimental Condensed Matter Physics ; Methods in Artificial Intelligence and Machine Learning
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 566291308

Over the past decade, advancements in synchrotron and neutron scattering experiments have led to unprecedented capabilities, including real-time and in operando studies with high spatial and temporal resolution. This progress, driven by enhanced source brilliance and faster detectors generating data at GB/s to TB/s rates, has outpaced traditional human-controlled experimental strategies and analysis tools, leading to inefficiencies in beamtime and energy resource utilization. Given the energy-intensive nature of large-scale facilities, such as the 44 GWh annual energy consumption at PETRA-III, optimizing the scientific return on invested energy is critical. This proposal addresses these challenges by integrating machine learning (ML) for autonomous "closed-loop" experimental strategies in synchrotron and neutron scattering experiments. ML will enable near real-time data analysis, feedback loops, and decision-making, enhancing efficiency in resource use and reducing human workload. By moving beyond traditional data analysis toward experiment control, ML-driven methodologies will enable autonomous experimentation, utilizing dynamic scans, sparse sampling, and adaptive feedback systems to optimize experimental out-comes. These measures will help to enhance resource-efficiency in terms of energy consumption, human resources including reduced travel, synchrotron and neutron beamtime, and ultimately a better use of the CO2 footprint of large-scale facilities. Our project initially targets the development of solutions for two specific science use cases: surface scattering techniques (including XRR, NR, and GIWAXS) and dynamic studies using X-ray photon correlation spectroscopy (XPCS). In these applications, ML has shown promise for parameter extraction, denoising, classification, and automated control. Nevertheless, substantial efforts are required to fully deploy these solutions on a large scale and achieve real-time optimization of beamline parameters. Building upon these advancements, we aim to extend the developed solutions to related use cases, facilitating the generalization and transfer of closed-loop feedback methodologies for synchrotron and neutron scattering experiments within the broader research community. This will involve adapting software frameworks to ensure compatibility and interoperability across multiple facilities. By creating modular, scalable systems and integrating solutions into existing control frameworks such as BlueSky and BLISS, the project will foster widespread adoption and streamlined implementation. This ambitious four-year initiative will build upon our strong preliminary ML work and our large network of collaborators to develop closed-loop solutions for our use cases. Serving as a pilot for broader applications, it will generalize workflows to accommodate other techniques and facilities, setting the foundation for future advancements in autonomous experimental strategies and resource-efficient research practices.

• DFG Programme: Further Instrumentation Related Funding