Speaker
Description
The Spallation Neutron Source (SNS) Superconducting Radio Frequency (SRF) Linac has been in production operation since 2006. Since that time much has been understood about causes for SRF downtime with a high-power proton beam. One of the important causes for downtime is related to repeated beam loss events which lead to the need to reduce cavity gradients to maximize reliability. Significant time has been spent to try to first reduce the frequency of beam loss events, and second to reduce the beam lost during each event. The need to reduce gradients has slowed significantly, but the need does remain. With the Proton Power Upgrade (PPU) ongoing which will double the beam power capability of the linac the need to further prevent beam loss events remains a high priority. The source and frequency of beam loss events are difficult to predict and prevent and the protection system turn off time is hardware limited. This all led to the idea to try to utilize machine learning to monitor beam pulses to try to predict an upcoming beam loss event and when predicted hold the beam off during the event to prevent beam loss occurring. The focus is not to prevent the cause of the beam loss but just prevent the beam from occurring in the SRF cavities. The timeline to reach the point of need for machine learning as well as the current implementation and future utilization will be discussed.
*ORNL is managed by UT-Battelle, LLC, under contract DE-AC05- 00OR22725 for the U.S. Department of Energy.
