Interpreting AI for Fusion: an application to Plasma Profile Analysis for Tearing Mode Stability

TL;DR

This paper introduces a physics-based interpretability framework for AI models predicting plasma instabilities in fusion reactors, validated through experiments and Shapley analysis to enhance trust and understanding.

Contribution

It presents a novel interpretability approach for AI in fusion, combining physics insights with ML predictions to improve safety and trustworthiness.

Findings

Core electron temperature and rotation are primary factors in TM stability.

Density profile has a light destabilizing effect.

Framework validated through DIII-D TM avoidance experiment.

Abstract

AI models have demonstrated strong predictive capabilities for various tokamak instabilities--including tearing modes (TM), ELMs, and disruptive event--but their opaque nature raises concerns about safety and trustworthiness when applied to fusion power plants. Here, we present a physics-based interpretation framework using a TM prediction model as a first demonstration that is validated through a dedicated DIII-D TM avoidance experiment. By applying Shapley analysis, we identify how profiles such as rotation, temperature, and density contribute to the model's prediction of TM stability. Our analysis shows that in our experimental scenario, a large density profile is lightly destabilizing, but core electron temperature and rotation peaking play the primary role in TM stability. This work offers a generalizable ML-based event prediction methodology, from training to physics-driven interpretability, bridging the gap between physics understanding and opaque ML models.