Title:Long Short-Term Memory Networks for Anomaly Detection in Magnet Power Supplies of Particle Accelerators

Abstract:This research introduces a novel anomaly detection method designed to enhance the operational reliability of particle accelerators - complex machines that accelerate elementary particles to high speeds for various scientific applications. Our approach utilizes a Long Short-Term Memory (LSTM) neural network to predict the temperature of key components within the magnet power supplies (PSs) of these accelerators, such as heatsinks, capacitors, and resistors, based on the electrical current flowing through the PS. Anomalies are declared when there is a significant discrepancy between the LSTM-predicted temperatures and actual observations. Leveraging a custom-built test stand, we conducted comprehensive performance comparisons with a less sophisticated method, while also fine-tuning hyperparameters of both methods. This process not only optimized the LSTM model but also unequivocally demonstrated the superior efficacy of this new proposed method. The dedicated test stand also facilitated exploratory work on more advanced strategies for monitoring interior PS temperatures using infrared cameras. A proof-of-concept example is provided.