Vertical instability prediction and its direction control using a support vector machine in integrated commissioning of JT-60SA solely based on magnetics

Abstract

We developed a redundant logic for predicting Vertical Displacement Events (VDEs) using only magnetic diagnostics with a Support Vector Machine (Inoue et al 2022 Nucl. Fusion 62 086007). This logic was experimentally validated in the integrated commissioning of JT-60SA. Triggering of VDEs results in significant asymmetric heat loads on first walls and electromagnetic loads on conducting materials. These concerns are mitigated by a newly proposed direction control for VDEs. We successfully detected VDEs in experiments and controlled their direction by setting the control voltage to zero upon detection, thus guiding VDEs in an intended direction and reducing the risk to the device from unmitigated VDEs in either direction, although completely avoiding VDEs remains the ideal. In the developed predictor, VDEs are predicted using proxies to monitor the controller’s performance with an adaptive voltage allocation scheme (Inoue et al 2021 Nucl. Fusion 61 096009). For future experiments, we plan to extend our approach to monitor more detailed information from the equilibrium controller, including control values of each PID component, which has been shown to enhance the prediction accuracy of VDEs.

Keywords

• vertical instability
• equilibrium control
• resistive wall mode
• machine learning
• support vector machine