Abstract:In the field of magnetic confinement plasma control, the accurate feedback of plasma position and shape primarily relies on calculations derived from magnetic measurements through equilibrium reconstruction or matrix mapping method. However, under harsh conditions like high-energy neutron radiation and elevated temperatures, the installation of magnetic probes within the device becomes challenging. Relying solely on external magnetic probes can compromise the precision of EFIT in determining the plasma shape. To tackle this issue, we introduce a real-time, non-magnetic measurement method on the HL-3 tokamak, which diagnoses the plasma position and shape via imaging. Particularly, we put forward an adapted Swin Transformer model, the Poolformer Swin Transformer (PST), to accurately and fastly interpret the plasma shape from the Charge-Coupled Device Camera (CCD) images. By adopting multi-task learning and knowledge distillation techniques, the model is capable of robustly detecting six shape parameters under disruptive conditions such as a divertor shape and gas injection, circumventing global brightness changes and cumbersome manual labeling. Specifically, the well-trained PST model capably infers R and Z within the mean average error below 1.1 cm and 1.8 cm, respectively, while requiring less than 2 ms for end-to-end feedback, an 80 improvement over the smallest Swin Transformer model, laying the foundation for real-time control. Finally, we deploy the PST model in the Plasma Control System (PCS) using TensorRT, and achieve 500 ms stable PID feedback control based on the PST-computed horizontal displacement information. In conclusion, this research opens up new avenues for the practical application of image-computing plasma shape diagnostic methods in the realm of real-time feedback control.

Plasma Real-time Diagnostics Based on Visual Image Recognition

Advances in Thermal Plasma Temperature Diagnostic Methods

Visible Image-Based Regression: A Novel Approach for Continuous and Real-Time Diagnosis of Plasma Molecular Temperatures

Real-Time Diagnosis of Plasma Molecular Temperature Based on OES and Elastic-Net Regression Analysis

Application of monochromatic image method in arc plasma diagnostics

Recognition of Ethylene Plasma Spectra 1D Data Based on Deep Convolutional Neural Networks

A Real-time Identification Method Based on Static-dynamic Optical Flow Features for Deep-sea Hydrothermal Plumes

A novel plasma optical emission spectroscopy method with neural network model for improving the accuracy of plasma diagnostics in low-temperature xenon plasma

Deep Learning for Plasma Tomography and Disruption Prediction from Bolometer Data

Recognition of Plasma Discharge Patterns Based on CNN and Visible Images

A Long Short-Term Memory Network for Plasma Diagnosis from Langmuir Probe Data

Infrared Image Defect Diagnosis Through LAB Space Transformation

Advancements in neural network techniques for electric and magnetic field reconstruction: Application to ion radiography

Chromatic-free Spatially Resolved Optical Emission Spectroscopy Diagnostics for Microplasma.

Low Temperature Plasma Diagnostics with Tunable Synchrotron Vacuum Ultraviolet Photoionization Mass Spectrometry.

Classification of the L-, H-mode, and plasma-free state: Convolutional neural networks and variational autoencoders on the edge reflectometer for KSTAR

Studies On Gas Concentration Reconstruction Methods For Simultaneous Temperature And H2o Concentration Tomographic Imaging In A Flame Based On Tunable Diode Laser Absorption Spectroscopy

A Spectral Clustering Based Method for Automatic Plasma Pattern Identification in Electron Cyclotron Emission Imaging on EAST Tokamak

Interpreting neural networks trained to predict plasma temperature from optical emission spectra

Experimental identification of Ion Cyclotron Emission on HL-2A using YOLO neural network algorithm

Adapted Swin Transformer-based Real-Time Plasma Shape Detection and Control in HL-3