Application of convolutional neural network to BEPCII SRF cavity fault analysis

Background The Superconducting Radio Frequency (SRF) cavity fault is one of the main factors limiting the availability of Beijing Electron Position Collider II (BEPCII) in the past several year. The traditional diagnostic methods for SRF cavity fault rely on manual intervention and empirical judgment, which are not only time-consuming but also susceptible to human bias, while the machine learning technology provides a new idea for SRF cavity fault diagnosis.
Purpose To improve the accuracy and processing speed of SRF cavity fault detection, the deep learning technology will be investigated in this paper.
Methods The Convolutional Neural Network (CNN) algorithms based on different network architectures were built for BEPCII SRF cavity fault classification, and the similarities and differences of them were analyzed and compared.
Results The study shows that CNN-based SRF cavity fault classification methods, especially ResNet-101 and AlexNet, can effectively distinguish SRF cavity faults. It exhibits high accuracy and generalization ability. It also provides an effective solution for SRF cavity fault diagnosis system.