Infrared measurement synthetic database for inverse thermography model based on deep learning

Abstract

Infrared (IR) diagnostics are used to measure the surface temperature of plasma-facing components in fusion devices. However, the interpretation of such images and any quantitative analysis are very complex in all-reflective environments because of unknown emissivity and multiple reflections. This is an important issue for safe operation and understanding of plasma-wall interaction phenomena. Moreover, it can also benefit other industrial applications e.g., monitor heating on production lines, additive manufacturing devices, etc. Recently, a first demonstration of simulation-assisted machine learning method has proven to be effective in retrieving surface temperatures from IR measurements. Yet, the performances of such an approach has been evaluated on a tokamak prototype considering pure specular surfaces. This paper presents an optimised generation of training dataset based on a synthetic data generated by a deterministic ray tracer model. The obtained results shows that the proposed methodology enables very accurate simulations minimizing the computational burden associate to greedy models (i.e., Monte Carlo type) to generate training data for realistic surfaces.