Demonstration of object location, classification, and characterization by developed deep learning dust ablation trail...

Area of Science:

• Fusion energy research
• Plasma physics
• Materials science in extreme environments

Background:

• Tokamak fusion reactors generate dust particles that can impact performance.
• Analyzing dust ablation trails is crucial for understanding plasma-wall interactions.
• Current methods for analyzing dust ablation trails are often manual and time-consuming.

Purpose of the Study:

• To develop and validate an automated deep learning code package for detecting and analyzing dust ablation trails in tokamaks.
• To benchmark the performance of the developed code package using plasma jet images.
• To demonstrate the applicability of the code package in real-world fusion experiments.

Main Methods:

• Development of a deep learning code package named Dust Ablation Trail Analysis (DATA).
• Training and testing the DATA package using 2440 plasma jet images, which mimic dust ablation clouds.
• Validation of the DATA package on dust ablation trails captured in the Experimental Advanced Superconducting (EAST) tokamak.

Main Results:

• The DATA package achieved 100% detection rate for plasma jets.
• Classification accuracy for plasma jets exceeded 99.9%.
• The package successfully analyzed dust ablation trails in the EAST tokamak, demonstrating high performance.

Conclusions:

• The developed DATA code package provides an effective automated solution for identifying and analyzing tokamak dust ablation trails.
• This tool significantly enhances the capacity for analyzing large datasets in fusion research.
• The DATA package facilitates further detailed investigations, including 3D reconstruction of dust and ablation phenomena.