Recent upgrades of synchrotron light source facilities towards ultra-low electron beam emittances allow increased photon beam brightness and coherence. New techniques for online modeling and control, taking advantage of modern Machine Learning approaches are required to fully utilize these new photon capabilities. We present recently developed reduced models for x-ray propagation that may enable an array of fast optimization methods for beamline alignment and reconfiguration. In particular, we have extended the analysis of the partially coherent Gaussian Schell model to include physical apertures and expressed it in terms of a Wigner function such that only second moment and centroid propagation is required. We have implemented this formalism within the SHADOW ray tracing code, providing fast, convenient transfer matrix computation down an x-ray beamline and subsequent moment propagation, including beam size, divergence and coherence properties. For the fully coherent case, we are developing tools for efficient Linear Canonical Transforms. On the optimization front, we have used Bayesian Optimization with Gaussian Processes and performed proof of principle automated alignment experiments on the Tender Energy Spectroscopy (TES) beamline at NSLS-II. These software tools are being integrated into the Sirepo web-based simulation framework as well as combined with the Bluesky control software suite in a dedicated package called Sirepo-Bluesky. We present an outlook on the progress we have made thus far, along with a future vision for this work.
The Sirepo-Bluesky library allows the performing of various types of Bluesky scans with Sirepo simulations acting as virtual beamlines and registration of the results with the Databroker library. We report on the progress made since the previous SPIE’2020. In particular, the support for Shadow3 and MAD-X simulation codes in Sirepo was added to the Sirepo-Bluesky library, and the API for the support of the Sirepo/SRW code was refactored. Significant efforts were put into reliable testing and documentation. A “digital twin” of the future NSLS-II ARI beamline was created and the future Bluesky scans were prototyped using the Sirepo/SRW simulations. This approach enables new optimization methods for automated instrument alignment based on the Ophyd/Bluesky and makes them transferable from simulated to various hardware backends.
The autonomous alignment of synchrotron beamlines is typically a high-dimensional, high-overhead optimization problem, requiring us to predict a fitness function in many dimensions using relatively few data points. A model that performs well under these conditions is a Gaussian process, upon which we can apply the framework of classical Bayesian optimization methods. We show that even with no prior data, a tailored Bayesian optimization algorithm is capable of autonomously aligning up to eight dimensions of a digital twin of the TES beamline at NSLS-II in only a few minutes. We implement this approach in a software package for automatic beamline alignment, which is available out-of-the-box for any facility that leverages the Bluesky environment for beamline manipulation and data acquisition.
Simulation of beamlines at light sources is an essential part of their design and commissioning. Such simulations can be performed by the Synchrotron Radiation Workshop (SRW) code, which now has a user-friendly, browser- based interface, Sirepo. The simulations, utilizing a concept of a "virtual" beamline, can aim to optimize the specific aspects of a beamline, such as maximization of the flux, minimization of the beam size, etc. These tasks are also performed at the physical beamlines using various alignment procedures. At NSLS-II these procedures are executed by the Bluesky data collection framework. The Sirepo-Bluesky interface leverages both systems to allow for the multiparameter optimization of the X-ray source and beamline optics with the power of bluesky's plans used for the daily experiments at NSLS-II, and databroker's capabilities to retrieve the captured data and metadata to perform further analysis. Such a "collaboration" between the frameworks can be used to store the simulated results in the same database as for the experimental data, and seamlessly apply the same analysis pipelines, demonstrated in recent publications. In a simulation, multiple parameters can be changed at once and be stored as a snapshot of the "virtual" beamline in time along with the corresponding results of the simulations. A global optimization algorithm (e.g., a genetic algorithm) can then utilize the data to find the best configuration for the desired outcome. Such an optimization procedure can be seamlessly applied to the real hardware by substituting the virtual motors and detectors by the real ones.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.