You are here

Adaptive Online Machine Learning for Dynamic Beam Diagnostics



b.      Adaptive Online Machine Learning for Dynamic Beam Diagnostics

Particle accelerators are large complex systems composed of hundreds to thousands of interconnected electro-magnetic components including radio frequency (RF) resonant accelerating structures for beam acceleration and longitudinal focusing and various magnets for beam steering and transverse focusing. Charged particle beams are themselves complex objects living in a six-dimensional phase space. They undergo complex collective effects such as coherent synchrotron radiation and vary with time in unpredictable ways. Sources of variation include accelerator RF phase and amplitude jitter, and magnet current jitter, and time-varying laser intensities and photoemission at the beam source. As bunches become shorter and more intense, the effects of intra-bunch collective interactions such as space charge forces and bunch-to-bunch influences such as wakefields also increase. Short, intense bunches are extremely difficult to accurately image because their dimensions are beyond the resolution of existing diagnostics and they may be destructive to intercepting diagnostics.


Proposals are sought for the design and implementation of adaptive machine learning methods as applied to time-varying systems since they have the potential to aid in the diagnostics and control of high-intensity, ultrashort beams by interfacing online models with real time non-invasive beam data, and thereby provide a detailed virtual view of intense bunch dynamics. The goal is to enable beam prediction and control, and to develop new diagnostics in order to increase beam phase-space density by at least an order of magnitude than currently achievable.


(See References 1 through 6 for further information.)


Questions – Contact: John Boger,

US Flag An Official Website of the United States Government