Infrared (IR) cameras are key diagnostics to monitor and control plasma-facing components (PFCs) in fusion devices. Nevertheless, the
use of all-metallic PFCs with low and variable emissivity (ε ~ 0.1-0.5) makes it difficult to obtain a correct surface temperature
measurement. The radiance collected by the IR camera includes both the thermal radiation emitted by the target and parasitic
radiation coming from the target's surroundings. Furthermore, target emissivity changes with the surface temperature and
roughness. This causes significant errors in the surface temperature measurement that we need to address for achieving high power
and safe plasma operation. Inaccurate interpretation of IR temperature measurement could endanger machine safety (temperature
underestimation) or on the contrary, lead to unnecessary pulse interruptions that reduce the overall performance of the machine
(temperature overestimation). The current approach is to convert the radiance collected by each pixel in an apparent using a physical
relationship between emitted radiance and temperature (Planck’s law). Because a portion of the collected radiance is due to reflection
on the target, a systematic error is made. This method is unable to recover the portion of reflected radiance and to deduce the correct
temperature of the target using only the emitted radiance. The proposed technique in this project is to use a digital twin of the
machine and to make in it assumptions on the temperatures of the different elements of the machine in order to simulate the images
that would be measured under these supposed conditions. By comparing to the real image measured by the camera and using
optimization techniques, the assumptions on the parameters can be updated until the simulated and measured images are close
enough. It is then deduced that the assumptions made on the parameters of the digital twin are correct and that they correspond to
the real temperatures in the machine.
