SRF Cavity Fault Classification Using Machine Learning At CEBAF

Abstract

The Continuous Electron Beam Accelerator Facility (CEBAF) at Jefferson Lab is the first large high power CW recirculating electron accelerator to make use of SRF accelerating structures. The structures are configured in two antiparallel linacs connected by arcs. Each linac consists of twenty C20/C50 cryomodules each containing eight 5-cell cavities and five C100 upgrade cryomodules each containing eight 7-cell cavities. Accurately classifying the source of cavity faults is critical for improving accelerator performance. A cavity fault triggers a waveform acquisition process where 17 waveform records sampled at 5 kHz are recorded for each of the 8 cavities in the affected cryomodule. The waveform record length is sufficiently long for transient microphonic effects to be observable. This data combined with archived signals sampled at 10 Hz are used to classify faults. Significant time is required for a subject matter expert to analyze and identify the intra-cavity signatures of imminent faults. This paper describes a path forward that utilizes machine learning for automatic fault classification. Post-training identification of the physical origins of faults are discussed, as are potential machinetrained model-free implementations of trip avoidance procedures. These methods should provide new insights into cavity fault mechanisms and facilitate intelligent optimization of cryomodule performance. DEFINITION OF THE PROBLEM The 12 GeV Upgrade for CEBAF was completed in September 2017. The project doubled the beam energy of the existing accelerator. To meet this energy goal, eleven new 100 MV cryomodules (called C100s) and RF systems were installed in 2013 (see Fig. 1) [1]. Currently the largest contributor to CEBAF downtime are beam trips caused by SRF cavities. During the last year there were an average of 6 RF trips an hour, accounting to roughly 15% of lost beam time per hour every day. To reduce the trip rate accelerating gradient of the cavity needs to be lowered, which means energy reach of CEBAF suffers. The cavities in a C100 cryomodule have strong cavity to cavity mechanical coupling. When one cavity trips off, the Lorentz force detuning causes vibrations in the cavity string that are sufficient to trip other cavities. In order to avoid trips, the entire string is switched to self-excited loop mode (frequency tracking) when one of the cavities trips and others become unstable. This is also the default response for various other off normal conditions, which makes it difficult to determine which cavity initiated the cascade of faults [2]. When a cavities trips off, it disrupts delivery of the beam to the experimental halls. Correctly classifying which of several known fault mechanisms caused the cavity to trip provides valuable information to control room operators on how to treat the offending cavity and ultimately helps to maintain greater beam availability to users [3]. Figure 1: Schematic of the CEBAF accelerator showing the locations of the 11 C100 cryomodules from which cavity fault data is recorded. USING THE FAULT IDENTIFICATION AND MACHINE OPERATION Some examples that illustrate how prompt identification of fault types can be useful in machine operation:  Fast Quenches: Identification of prompt “quenches” of cavities where the stored energy in the cavity is dissipated in times that are much shorter than is possible due to thermal quenches. These events were identified in the CEBAF operation as a gas discharge inside the cavity where stored energy is transferred to electrons produced by the discharge in times on the order of 10 μs. When these types of events occur in either the first or last cavity in the cryomodule there is pressure outburst observed in the beam line ion pump. In some cryomodules these events started occurring multiple times per day after weeks of no events and at gradients well below previously determined quench gradients. This can indicate gas loading in the beamline or the warm-to-cold transition in the RF waveguides. In addition to the temporary mitigation of reducing the gradient, identifying this type of fault can indicate a vacuum problem or the need to thermally cycle the cryomodule. ___________________________________________ * Authored by Jefferson Science Associates, LLC under U.S. DOE Contract No. DE-AC05-06OR23177 † shabalin@jlab.org 10th Int. Partile Accelerator Conf. IPAC2019, Melbourne, Australia JACoW Publishing ISBN: 978-3-95450-208-0 doi:10.18429/JACoW-IPAC2019-TUXXPLM2 MC7: Accelerator Technology T07 Superconducting RF TUXXPLM2 1167 Co nt en tf ro m th is w or k m ay be us ed un de rt he te rm so ft he CC BY 3. 0 lic en ce (© 20 19 ). A ny di str ib ut io n of th is w or k m us tm ai nt ai n at tri bu tio n to th e au th or (s ), tit le of th e w or k, pu bl ish er ,a nd D O I