Temperature Estimation in Fusion Devices Using Machine Learning Techniques on Infrared Specular Synthetic Data

Infrared (IR) imaging systems are common diagnostics for monitoring in-vessel components in the rmonuclear fusion devices (tokamak). Nevertheless, IR interpretation in fully metallic environment is complex due to the presence of multiple reflections and the change of optical properties of materials as the fusion operation progresses. This causes high errors on the surface temperature measurement which is a risk for machine protection. The paper presents a first demonstration of simulation-assisted machine learning method for retrieving the surface temperature from IR measurement on metallic targets with unknown properties. The technique relies on the training of a convolutional neural network on a synthetic dataset generated by a deterministic ray tracer. The performances of such an approachis first proven on tokamak prototype considering pure specular surfaces.