Data-Driven Gradient Optimization for Field Emission Management in a Superconducting Radio-Frequency Linac

Abstract

Field emission can cause significant problems in superconducting radio-frequency linear accelerators (linacs). When cavity gradients are pushed higher, radiation levels within the linacs may rise exponentially, causing degradation of many nearby systems. This research aims to utilize machine learning with uncertainty quantification to predict radiation levels at multiple locations throughout the linacs and ultimately optimize cavity gradients to reduce field emission induced radiation while maintaining the total linac energy gain necessary for the experimental physics program. The optimized solutions show over 40% reductions for both neutron and gamma radiation from the standard operational settings.

Figures (6)

• Figure 1: NDX 2L25 detector positioned next to the beamline connecting two cryomodules at a distance of about 0.5m. Neighbouring cryomodules are separated by a 1.2m long beamline. The white polyethylene outer layer of the detector is yellowing due to FE radiation.
• Figure 2: Layout of the CEBAF south linac. Cryomodules are outlined in blue, detectors in red. There are 8 individual cavities per cryomodule (shown in black). The south linac contains three cyromodule types indicated by their background color: C25, C50, and C100, where the numbers indicate the nominal energy gain in MeV expected from each type.
• Figure 3: Example of data drift. PCA-based dimensionality reduction of measured cavity gradients from the different data sets we collected. (S) and (D) denote the scan and demonstration datasets, respectively.
• Figure 4: Radiation delta from the initial timestep for observations (in blue) and model responses (in orange) on the May 14 dataset. Uncertainties are shown as an orange shaded region equivalent to $\pm 3 \sigma$. By plotting the radiation delta rather than the exact observations and predictions, we remove any systematic shift and highlight true missed predictions. While most model responses are well correlated, we see a possible change of behavior for 2L24 (near 10:25).
• Figure 5: Delta between gradient settings chosen by our NSGA optimizer and the baseline settings (in blue). Since a few settings were modified between the baseline and the beginning of the demonstration, the true demonstration deltas (for \ref{['fig:may19-radiation-responses']}) are shown as red lines. R221 was turned off after our baseline was taken, thus this bar is replaced by an "x". The demonstration delta for R2Q7 was +3.8 (not shown).