Study on the State Prediction of a Pool-Cooled Large Superconducting Coil Using Machine Learning

Aug 2, 2026

Upload by Oliver

A machine learning approach is developed to improve reliability of the LHD subcooling system, which consists of pool-cooled large superconducting coils wound with NbTi superconductors. The state-prediction model is trained using measurement data accumulated during the LHD plasma experimental campaign. It enables forecasting temperature changes in the system caused by coil excitation and discharging. For the typical trapezoidal helical coil current waveform, the model provides high prediction accuracy compared with observed measurements.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

Introduction

LHD Subcooling System for Helical Coils

Machine Learning Model

Model Development

Validity and Comparison

Study on the State Prediction of a Pool-Cooled Large Superconducting Coil Using Machine Learning | メタデータ | 言語: English 出版者: IEEE 公開日: 2024-08-01 キーワード: Predictive models, Machine learning, Temperature measurement, Training data, Data models, Current measurement, Plasma temperature 作成者: OBANA, Tetsuhiro メールアドレス: 所属: | | --- | --- | | URL | [http://hdl.handle.net/10655/0002000749](http://hdl.handle.net/10655/0002000749) | Study on the state prediction of a pool-cooled large superconducting coil using machine learning T. Obana Abstract—For the LHD subcooling system composed of poolcooled large superconducting coils wound with NbTi superconductors, a machine learning technique was introduced to increase the reliability of the system. The machine learning model for the state prediction of the system was developed using the technique, together with the data accumulated in the LHD plasma experimental campaign. Regarding the temperature changes in the system due to coil excitation and discharging, it is possible to make predictions using the model. Especially for the usual coil current waveform ina helical coil operation, which is a trapezoidal waveform, the model achieved high prediction accuracy. Index Terms—Machine learning, state prediction, pool-cooled superconducting coil, NbTi superconductor, Helical coil, Subcooled helium, Long short-term memory (LSTM) I. INTRODUCTION INeqheipmsupenterconducof a fusinngemagnexperimsysntaltemdev,iwceh, hlasgeteucmobreer of measurement sensors are installed in the system for its operation and monitoring. Although each measurement sensor is inspected and repaired during the maintenance period of the experimental device, there is a concern that trouble may occur in a sensor that has been used for a long period of time. If trouble occurs in a sensor that is important for system operation, it is necessary to stop the plasma experiment, raise the temperature of the superconducting magnet system to room temperature, and repair or replace the measurement sensor. In this study, the development of a machine learning model was conducted to predict the state of the superconducting magnet system, based on the measurement data accumulated during system operation. By using the model, even if trouble occurs in the measurement sensor, the system operation is continued. Namely, the reliability of the superconducting magnet system is increased with the machine learning model. For example, the ITER superconducting magnet system, which is currently under construction, is also being modeled using machine learning technology [1, 2] . The object of model development in this study is the subcooling system for the Large Helical Device (LHD) [3, 4] . This system is suitable for the model development because an amount of measurement data, which is used as training data Manuscript receipt and acceptance dates will be inserted here.(Corresponding author: Tetsuhiro Obana.) T. Obana is with the National Institute for Fusion Science, 509-5292, Japan (e-mail: [obana.tetsuhiro@nifs.ac.jp](obana.tetsuhiro@nifs.ac.jp)). for the modeling, has already been accumulated through the LHD plasma experimental campaign. This paper describes the details of the modeling and the validity of the machine learning model, while comparing predicted values from the model with the measurement results in the subcooling system. Additionally, the difference between the machine learning model and the physics model, which have been developed in a previous study, is discussed. Fig. 1. Flow diagram ofthe LHD subcooling system. II. LHD SUBCOOLING SYSTEM FOR HELICAL COILS The LHD is a plasma experimental device composed of superconducting helical coils [5, 6] . The helical coils are a pair of pool-cooled superconducting coils wound with superconductors which consist of NbTi/Cu strands, a pure aluminum stabilizer and a copper housing [7, 8] . Also, the coils are composed of three coil winding blocks called H-I, H

FAQs

Why is sensor reliability critical in the LHD subcooling system?

Because measurement sensors are required for operation and monitoring, a malfunction in an important sensor can force stopping plasma experiments and repairing or replacing the sensor.

How does the machine learning model improve system reliability?

It predicts the superconducting magnet system state from accumulated operating measurements, allowing continued operation even when a measurement sensor trouble occurs.

What inputs and data are used to build the state prediction model?

The model is trained using data accumulated during the LHD plasma experimental campaign, targeting temperature changes caused by coil excitation and discharging.

Aug 2, 2026

Upload by Oliver