Anomaly Detection in Particle Accelerators using Autoencoders

TL;DR

This paper explores the use of autoencoders, a type of neural network, for detecting magnet faults in particle accelerators, addressing challenges like data imbalance and lack of labels.

Contribution

It demonstrates the application of autoencoder reconstruction analysis for predicting magnet faults in the APS storage ring, advancing anomaly detection methods in accelerators.

Findings

Autoencoders effectively identify magnet faults in accelerator data.

Reconstruction errors correlate with fault occurrences.

The method addresses data imbalance and unlabeled data challenges.

Abstract

The application of machine learning techniques for anomaly detection in particle accelerators has gained popularity in recent years. These efforts have ranged from the analysis of quenches in radio frequency cavities and superconducting magnets to anomalous beam position monitors, and even losses in rings. Using machine learning for anomaly detection can be challenging owing to the inherent imbalance in the amount of data collected during normal operations as compared to during faults. Additionally, the data are not always labeled and therefore supervised learning is not possible. Autoencoders, neural networks that form a compressed representation and reconstruction of the input data, are a useful tool for such situations. Here we explore the use of autoencoder reconstruction analysis for the prediction of magnet faults in the Advanced Photon Source (APS) storage ring at Argonne National Laboratory.