Exploring NAS for anomaly detection in superconducting cavities of particle accelerators

The European X-Ray Free Electron Laser is the largest particle accelerator for X-ray laser generation worldwide. To ensure a safe and efficient operation, the plant uses various monitoring systems, especially in the linear accelerator. The low-level radio frequency system has shown reliability in diagnostics, particularly in quench detection. A quench refers to a superconducting radio frequency cavity losing its superconductivity and possibly causing a downtime. The diagnostics solution, however, can be enhanced in terms of robustness and functionality. Currently, the focus is on integrating artificial intelligence to improve quench identification. Thus, a lightweight machine learning-assisted approach targeting FPGA deployment is developed. It relies on the augmentation of a physical model-based anomaly detection approach with neural network models to distinguish the quenches from the other anomalies. This paper presents the solution in which neural architecture search is applied, and elaborates on how visualizing and analyzing the anomaly detection results can provide critical insights for both short-term diagnostics and long-term pattern identification.