Beam-based rf station fault identification at the SLAC Linac Coherent Light Source

Abstract

Significant advances in utilizing deep learning for anomaly detection have been made in recent years. However, these methods largely assume the existence of a normal training set (i.e., uncontaminated by anomalies) or even a completely labeled training set. In many complex engineering systems, such as particle accelerators, labels are sparse and expensive; in order to perform anomaly detection in these cases, we must drop these assumptions and utilize a completely unsupervised method. This paper introduces the Resilient Variational Autoencoder (ResVAE), a deep generative model specifically designed for anomaly detection. ResVAE exhibits resilience to anomalies present in the training data and provides feature-level anomaly attribution. During the training process, ResVAE learns the anomaly probability for each sample as well as each individual feature, utilizing these probabilities to effectively disregard anomalous examples in the training data. We apply our proposed method to detect anomalies in the accelerator status at the SLAC Linac Coherent Light Source (LCLS). By utilizing shot-to-shot data from the beam position monitoring system, we demonstrate the exceptional capability of ResVAE in identifying various types of anomalies that are visible in the accelerator.

Robust errant beam prognostics with conditional modeling for particle accelerators

“…Recently, VAEs have been applied for high voltage converter modulator anomaly detection at SNS accelerator [34,35]. Methods have also been proposed, such as [8], to deal with impurities in the normal training data for VAEs.…”

“…Existing accelerator data provides a large amount of normal samples and a small number of anomaly samples. As such, researchers have been resorting to semi-supervised learning methods such as auto-encoders (AE) [5] or variational AE (VAE) [6] to leverage a large amount of existing normal data samples [7][8][9]. AE and VAE models learn to reproduce the normal data samples and predict anomalies based on the reconstruction error (the error between the input data and the corresponding reproduced data).…”

Coincident learning for unsupervised anomaly detection of scientific instruments

“…We now return to our original motivating problem of identifying the source of RF station faults at LCLS. We utilize the dataset assembled and described in [32,61]. The subsystem (s) data input consists of time-series data for a single RF station, with one data point approximately every 5 s. We use a sensitive trigger to actively select time windows with the possibility of an event to reduce data requirements; any relative change of 0.5% will trigger a window to be acquired.…”

“…To identify RF station faults, we have two data inputs, one containing data from an RF station subsystem (s) and one containing electron energy data beam-position monitors (BPMs) that monitor electron beam quality (q). The task is described in detail in [32]. During normal operation, the variability in the signals is independent (i.e.…”

A smart alarm for particle accelerator beamline operations

“…These algorithms work by identifying unpermitted deviations from acceptable, usual or standard conditions in an automated way [2]. Given that radio-frequency (RF) cavities are the fundamental building blocks of particle accelerators, and given that these devices generate information-rich data, a lot of research has been directed toward detection, isolation, classification, and prediction of anomalies in RF systems [3][4][5][6]. Recent work also applies anomaly detection methods to superconducting magnets [7], to identify and remove malfunctioning beam position monitors (BPMs) [8], and classify or predict errant signals [9,10], among many other applications [11][12][13][14][15].…”