Artificial Neural Networks for Real-time Diagnostic of High-Z Impurities in Reactor-relevant Plasmas

The operation of JET with a new wall, made of beryllium in the main chamber and a tungsten divertor, will require additional care in handling plasma-wall interactions, since these new materials are certainly much less forgiving than the present ones. In particular, detecting tungsten will be extremely important not only for safety but also to understand the behaviour of high-Z impurities in reactor-relevant plasmas. In this paper Artificial Neural Networks are investigated to face the problem of real-time detection of high-Z impurities in plasma scenarios of ITER relevance. The data were collected with JET spectroscopy in a series of experiments, where laser blow-off was used to inject the various impurities. A wide range of plasma parameters was explored to cover the most important regions of the spectra. The good results obtained in recognizing the most important lines of the relevant materials prove that Artificial Neural Networks are strong candidates for real-time monitoring of the impurities both for protection purposes and for investigation of first-wall erosion.