AI/ML-Enabled Wavefront Sensing and Phase Engineering for Next-Generation Synchrotron Beamlines
ORAL · Invited
Abstract
Autonomous beamline control and experiment optimization are key objectives for next-generation synchrotron facilities. At the Advanced Photon Source (APS), following its recent upgrade to the world’s brightest synchrotron source, the increased brightness and coherence place stringent demands on wavefront quality, stability, and adaptive control. Meeting these requirements calls for tightly integrated diagnostics and AI-driven control, supported by physics-informed digital tools that link source, optics, and detector behavior.
We present recent progress in AI/ML-based X-ray wavefront sensing and phase engineering for intelligent beam control at APS. On the diagnostics side, we develop coded-mask-based wavefront sensing combined with a neural-network phase reconstruction framework, SPINNet. Originally developed for phase contrast imaging, SPINNet is extended to wavefront metrology by incorporating coded-mask-enabled simulated reference patterns, enabling absolute phase reconstruction from single-shot measurements. Retrieved wavefront phases are used directly, or via backpropagation, to extract beam properties and quantify optical aberrations.
These diagnostics enable automated optics alignment and adaptive wavefront control. We demonstrate AI-assisted autofocusing and phase optimization at APS using bendable Kirkpatrick–Baez mirrors, closing the loop between wavefront sensing and optics control. Ongoing work extends this approach to nanofocusing optics optimization at the APS In Situ Nanoprobe beamline, targeting focal spot sizes approaching 20 nm. For finer phase manipulation, we develop advanced ML-based control strategies, including tandem neural networks, to drive bimorph adaptive mirrors with distributed actuators for high-dimensional wavefront engineering.
We conclude with perspectives on integrating these capabilities with experiment-level performance metrics, where physics-based digital twins serve as enabling tools for robust AI training and closed-loop optimization from source to detector.
We present recent progress in AI/ML-based X-ray wavefront sensing and phase engineering for intelligent beam control at APS. On the diagnostics side, we develop coded-mask-based wavefront sensing combined with a neural-network phase reconstruction framework, SPINNet. Originally developed for phase contrast imaging, SPINNet is extended to wavefront metrology by incorporating coded-mask-enabled simulated reference patterns, enabling absolute phase reconstruction from single-shot measurements. Retrieved wavefront phases are used directly, or via backpropagation, to extract beam properties and quantify optical aberrations.
These diagnostics enable automated optics alignment and adaptive wavefront control. We demonstrate AI-assisted autofocusing and phase optimization at APS using bendable Kirkpatrick–Baez mirrors, closing the loop between wavefront sensing and optics control. Ongoing work extends this approach to nanofocusing optics optimization at the APS In Situ Nanoprobe beamline, targeting focal spot sizes approaching 20 nm. For finer phase manipulation, we develop advanced ML-based control strategies, including tandem neural networks, to drive bimorph adaptive mirrors with distributed actuators for high-dimensional wavefront engineering.
We conclude with perspectives on integrating these capabilities with experiment-level performance metrics, where physics-based digital twins serve as enabling tools for robust AI training and closed-loop optimization from source to detector.
*This work was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under contract DE-AC02-06CH11357.
–
Presenters
-
Xianbo Shi
- Advanced Photon Source - Argonne National Laboratory