Thermal damage detection of EAST internal component based on machine learning
Open Access
- 29 May 2025
- journal article
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 6 (2) , 025045
Abstract
During the discharge process, thermal events are common and destructive in Tokamak experiments. High energy particles in plasma can collide with device components such as limiters and diverters, which can lead to overheating, material cracking, and even damage to the structure of the device. Therefore, we need efficient detection methods to monitor thermal events in real-time. To address the challenge of identifying thermal damage to internal components of the first wall during the Experimental Advanced Superconducting Tokamak (EAST) discharges, we introduce the YOLOv8 model for hotspot detection on EAST and present the customized EAST- You Only Look Once (YOLO) algorithm, derived from an enhanced YOLOv8 framework. YOLOv8, known for its strong performance in real-time object detection, serves as a robust base model for this task. However, its performance on small object detection, such as early-stage thermal damage, is limited. Improvements are needed for specialized tasks, particularly for early warning and precise identification of small internal component damage during the initial stages of EAST discharges. we enhance the YOLOv8 model by incorporating specialized layers for detecting small targets and integrating the CBAM attention mechanism. These adjustments result in a network model capable of sensitively detecting internal component damage in EAST. Experimental results demonstrate that EAST-YOLO surpasses several versions of traditional YOLOv8 models in model evaluation metrics, achieving a precision of 97.5%, mAP50 of 97.7%, and a Recall of 94.0%. This problem- oriented approach significantly improves the operational safety and stability of the EAST device by enabling early detection of thermal events. The AI-based detection method provides a new solution to safeguarding the fusion device, while also offering potential avenues for integrating artificial intelligence technologies into EAST feedback control and operating systems in the future.Keywords
Funding Information
- The Strategic Priority Research Program of Chinese Academy of Sciences (No. XDB0790101)
- Anhui Provincial Major Science and Technology Project (2023z020004)
- the youth innovation promotion association cas (No. 2023470)
- The National Natural Science Foundation of China (Nos.12375230)
- The National MCF Energy R&D Program (2019YFE03040000)
- the youth innovation promotion association cas (No. 2023470)
This publication has 23 references indexed in Scilit:
- YOLO-BS: a traffic sign detection algorithm based on YOLOv8Scientific Reports, 2025
- Enhanced object detection in pediatric bronchoscopy images using YOLO-based algorithms with CBAM attention mechanismHeliyon, 2024
- Testing of ITER-grade plasma facing units in the WEST tokamak: Progress in understanding heat loading and damage mechanismsNuclear Materials and Energy, 2023
- Deep learning and image processing for the automated analysis of thermal events on the first wall and divertor of fusion reactorsPlasma Physics and Controlled Fusion, 2022
- Diagnostics of the dynamics of material damage by thermal shocks with the intensity possible in the ITER divertorPhysica Scripta, 2018
- Development of laser-based technology for the routine first wall diagnostic on the tokamak EAST: LIBS and LIASPhysica Scripta, 2017
- Wide-angle ITER-prototype tangential infrared and visible viewing system for DIII-DReview of Scientific Instruments, 2014
- ITER transient consequences for material damage: modelling versus experimentsPhysica Scripta, 2007
- Thermal stability of the tokamak plasma edgePlasma Physics and Controlled Fusion, 1997
- The purpose, status and future of fusion researchPlasma Physics and Controlled Fusion, 1993