Link do spotkania w aplikacji Microsoft Teams: https://tiny.pl/khqry7hn

Abstract

Monitoring of fuel retention and chemical composition of the first wall and divertor are one of the priorities in plasma-wall interaction research for future development of the next-step tokamaks as ITER. Being a remote and typically contactless method, LIBS (Laser Induced Breakdown Spectroscopy) has emerged as a candidate for this purpose. To verify its performance in reactor conditions, in-situ experiment with LIBS on remote arm has been performed on the Joint European Torus (JET) device. The international collaboration in this experiment involved participants from Finland, Germany, Italy, Estonia, Latvia, Poland, Slovakia and the UK.

The LIBS head was mounted on MASCOT, JET’s remotely operated robotic system designed for maintenance, repair and diagnostics tasks which enabled reaching over 800 locations inside the JET vessel. The signals were connected to a set of spectrometers with the use of 20 m long fiber cable.

The need of analysis of such massive number of spectra was a great opportunity for application of machine learning models which are under development and research at the IPPLM. Among other methods, autoencoders have been used in order to reduce data volume and reduce noises.

Besides of the details of the LIBS @JET experiment, this presentation will give insight into the deep learning models being under development at the IPPLM together with the introduction of fundamentals of machine learning and clarifications of the most often misconceptions in this topic.