In this paper, the possibility of using neural networks for fast tomographic reconstructions of tokamak plasma soft X-ray (SXR) emissivity is investigated. Indeed, the radiative cooling of heavy impurities like tungsten could be detrimental for the plasma core performances of ITER, thus developing robust and fast SXR diagnostic tools is a crucial issue to monitor the impurities and to mitigate in real-time their central accumulation. As preliminary work, a database of emissivity phantoms with associated synthetic measurements is used to train the neural network to solve the inversion problem. The inversion method, training process, and first tomographic reconstructions are presented with the perspectives about our future work.

In this work, two laser-produced plasma (LPP) sources – extreme ultraviolet (EUV) and a LPP soft X-ray (SXR) source were used to create Ne photoionized plasmas. A radiation beam was focused onto a gas stream, injected into a vacuum chamber synchronously with the radiation pulse. EUV radiation spanned a wide spectral range with pronounced maximum centered at λ≈11 nm, while in case of the SXR source spectral maximum was at λ≈1.4 nm. Emission spectra of photoionized plasmas created this way were measured in a wide spectral range λ = 10–100 nm. The dominating spectral lines originated from singly charged ions (Ne II) and neutral atoms (Ne I). For the highest radiation fluence, spectral lines originating from Ne III and even Ne IV species were detected. Differences between the experimental spectra, obtained for all irradiation conditions, were analyzed. They were attributed either to different fluence or spectral distribution of driving photons.