AUTHOR=Vidyaratne Lasitha , Carpenter Adam , Powers Tom , Tennant Chris , Iftekharuddin Khan M. , Rahman Md Monibor , Shabalina Anna S. 

TITLE=Deep Learning Based Superconducting Radio-Frequency Cavity Fault Classification at Jefferson Laboratory

JOURNAL=Frontiers in Artificial Intelligence

VOLUME=Volume 4 - 2021

YEAR=2022

URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2021.718950

DOI=10.3389/frai.2021.718950

ISSN=2624-8212

ABSTRACT=This work investigates the efficacy of deep learning for classifying C100 superconducting radio-frequency (SRF) cavity faults in the Continuous Electron Beam Accelerator Facility (CEBAF) at Jefferson Lab. CEBAF is a large, high-power continuous wave recirculating Linac that utilizes  418 SRF cavities to accelerate electrons up to 12 GeV. Recent upgrades to CEBAF include installation of 11 new cryomodules (88 cavities) equipped with a low-level RF system that records RF time-series data from each cavity at the onset of an RF failure. Typically, subject matter experts (SME) analyze this data to determine the type of fault and identify the cavity where it originated. This information is subsequently utilized to identify failure trends and to implement corrective measures on the offending cavity. This manual inspection process of large-scale, time-series data, generated by frequent system failures is tedious and time consuming. Consequently, this work explores the development of deep learning models to automate the process of cavity and fault classification. We examine two seminal deep learning (DL) architecture types for the cavity RF data analysis tasks: deep recurrent neural networks (RNN) and deep convolutional neural networks (CNN). RNNs are specifically designed to process time-series data in its raw form. However, CNNs are tailored to process multidimensional images. Therefore, we further investigate suitable transformations from time-series to image representations for the proposed CNN architectures. We provide a detailed analysis on the performance of individual models using an RF waveform dataset built using past operational runs of CEBAF. In particular, the performance of deep recurrent learning (DRL) models incorporating long-short term memory (LSTM) are analyzed along with CNN performance using two data representation methods developed for this study. Additionally, the efficacy of these DL models is also compared with a state-of-the-art fault and cavity identification system built using traditional machine learning (ML) methods based on the same dataset. The results demonstrates that the DL architectures obtain competitive cavity identification and sufficient fault classification performance with substantial inference speed gain, and is therefore a viable alternative to the current ML system.