Author: Pirayesh, R.
Paper Title Page
WEPLM01 Studies in Applying Machine Learning to Resonance Control in Superconducting RF Cavities 659
SUPLO01   use link to see paper's listing under its alternate paper code  
  • J.A. Diaz Cruz, S. Biedron, M. Martinez-Ramon, R. Pirayesh, S.I. Sosa Guitron
    University of New Mexico, Albuquerque, USA
  • J.A. Diaz Cruz
    SLAC, Menlo Park, California, USA
  Traditional PID, active resonance and feed-forward controllers are dominant strategies for cavity resonance control, but performance may be limited for systems with tight detuning requirements, as low as 10 Hz peak detuning (few nanometers change in cavity length), that are affected by microphonics and Lorentz Force Detuning. Microphonic sources depend on cavity and cryomodule mechanical couplings with their environment and come from several systems: cryoplant, RF sources, tuners, etc. A promising avenue to overcome the limitations of traditional resonance control techniques is machine learning due to recent theoretical and practical advances in these fields, and in particular Neural Networks (NN), which are known for their high performance in complex and nonlinear systems with large number of parameters and have been applied successfully in other areas of science and technology. In this paper we introduce NN to resonance control and compare initial performance results with traditional control techniques. An LCLS-II superconducting cavity type system is simulated in an FPGA, using the Cryomodule-on-Chip model developed by LBNL, and is used to evaluate machine learning algorithms.  
DOI • reference for this paper ※  
About • paper received ※ 05 September 2019       paper accepted ※ 15 September 2019       issue date ※ 08 October 2019  
Export • reference for this paper using ※ BibTeX, ※ LaTeX, ※ Text/Word, ※ RIS, ※ EndNote (xml)