SBIR/STTR Award attributes
There are more than 100 x-ray beamlines in the US each producing radiation that is used to study many classes of experimental samples with a wide array of experimental techniques. Each beamline is tailored specifically for the experimental methods used at the end-station, and is designed to optimize a variety of different photon beam properties. During the design phase of a beamline, a wide array of tools are used to ensure that it will achieve its specifications. Once the beamline is built and operating however, the modeling tools are not frequently used. Additionally, for beam-line tuning typically is accomplished with tools such as manual tuning, save-states, classical control methods, and look-up tables used for the beamline tuning and control. In this project, we will develop online modeling methods using reduced models and machine learning for automated beamline control. Physics models will be fast and yet still capture the properties of interest in the photon beam. Using these fast-executing models, we will apply machine learning to train neural- network controllers for fast reconfiguration of beam-lines for users. During Phase I we will demonstrate the use of reduced physics-based models for simulating x-ray beamlines and benchmark these models against high fidelity simulations of the beamlines. We will also demonstrate the use of machine learning for automatic reconfiguration of the beamline and prototype a graphical user interface for these online models. X-ray synchrotron radiation light sources are engines for scientific discovery in condensed matter physics, biology, chemistry, and medicine. Increasing scientific production further can be accomplished by minimizing the time spent reconfiguring the optical beamlines. Additionally, the ability to rapidly optimize beamlines for new experiments will increase their scientific output. The commercialization strategy is to provide highly skilled scientific and engineering R&D consulting across a broad range of scientific submarkets. We also plan to sell support subscriptions to our web-based scientific software tools.
