Predicting LTS Superconducting NbTi Magnet Quench: A 1D-CNN Model for Real-Time Implementations
IEEE Transactions on Applied Superconductivity · 2026-08 · 36 (5) · 1-5
DOI: 10.1109/tasc.2026.3680889 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Applied Superconductivity, 2026-08. Verfasst von Aline Y. Machado und 2 weiteren Autor:innen. Die Arbeit nennt 19 Literaturangaben.
Abstract (Originalsprache)
In superconducting systems, a critical challenge is managing spontaneous quench events, where a part of the superconducting material transitions to a normal resistive state and, if uncontrolled, can cause severe damage. This work presents a machine learning approach for a Quench Detection System (QDS), designed for the Superconducting Wavelength Shifter (SWLS) project at the Sirius particle accelerator, located at the Brazilian Center for Research in Energy and Materials (CNPEM). We developed a light-weight quench prediction system based on a Convolutional Neural Network (CNN) to forecast voltage increases across the Low-Temperature Superconducting (LTS) coils. We gathered voltage signals from quench training experiments using a vertical cryostat test setup, then we preprocessed the dataset of 43 experiments by characterizing and modeling the setup noise and applied signal processing techniques to prepare the dataset. Furthermore, we trained, tested, and evaluated a 1D CNN classifier to predict quench occurrences. The model met the performance requirements for real-time implementation on the Sirius's SWLS, predicting quench events with a lead time around 10 milliseconds. The network achieved 96% accuracy, 100% precision, 78% recall, and a 0.88 F1 score, a sufficient performance to be used as an early-warning complementary protection layer. It features a minimalist architecture with three layers and 525 parameters, making it suitable for implementation on the project's ZYNQ-based FPGA. This work provides a compact CNN model and generalizable preprocessing steps, offering a solution that is suitable for real-time deployment, enhancing operational safety, and adding a protection layer for the superconducting elements of the Sirius's SWLS.
Zitieren
APA
Machado, A. Y., Trindade, P. H. A., & Martinez, A. C. (2026). Predicting LTS Superconducting NbTi Magnet Quench: A 1D-CNN Model for Real-Time Implementations. IEEE Transactions on Applied Superconductivity, 36(5), 1-5. https://doi.org/10.1109/tasc.2026.3680889
Vancouver
Machado AY, Trindade PHA, Martinez AC. Predicting LTS Superconducting NbTi Magnet Quench: A 1D-CNN Model for Real-Time Implementations. IEEE Transactions on Applied Superconductivity. 2026;36(5):1-5. doi:10.1109/tasc.2026.3680889
BibTeX
@article{Machado2026,
author = {Machado, Aline Y. and Trindade, Pedro H. A. and Martinez, Amanda C.},
title = {Predicting LTS Superconducting NbTi Magnet Quench: A 1D-CNN Model for Real-Time Implementations},
journal = {IEEE Transactions on Applied Superconductivity},
year = {2026},
volume = {36},
number = {5},
pages = {1--5},
doi = {10.1109/tasc.2026.3680889}
}
EUVIMED