Using convolutional neural networks to detect edge localized modes in DIII-D from Doppler backscattering measurements

Area of Science:

• Plasma physics
• Fusion energy research
• Machine learning applications

Background:

• Edge localized modes (ELMs) are plasma ejections in H-mode tokamak operation.
• ELMs cause energy loss and vessel damage, necessitating effective detection and mitigation strategies.
• Current diagnostic methods like deuterium-alpha (Dα) spectroscopy have limitations.

Purpose of the Study:

• To develop and evaluate a neural network model for detecting ELMs using Doppler backscattering (DBS) data.
• To establish DBS as a viable diagnostic for ELM detection in future operational tokamaks.
• To demonstrate the broader applicability of neural networks to DBS diagnostic data.

Main Methods:

• Utilized the DIII-D tokamak database for training and testing.
• Trained a neural network to classify time steps based on ELM events using Dα data as ground truth.
• Evaluated model performance across various ELM types and confinement regimes.

Main Results:

• Achieved a high f1-score of 0.93 on test data, indicating robust ELM detection.
• Demonstrated consistent performance across different ELM regimes (grassy, RMP mitigated, wide-pedestal).
• Confirmed the broad applicability of neural networks for analyzing DBS diagnostic data.

Conclusions:

• Neural networks provide a promising and accurate method for ELM detection using DBS data.
• DBS, with its high temporal resolution and robustness, is a valuable diagnostic for future tokamaks.
• This work validates the use of machine learning for processing DBS data, extending beyond ELM detection.