The JET Alfvén Eigenmode Local Manager for the real-time detection and tracking of MHD instabilities

Abstract

In this work we report the successful application of an innovative method, based on the Sparse Representation of signals, to perform a real-time, unsupervised detection of the individual components in a frequency degenerate, multi-harmonic spectrum, using a small number of data un-evenly sampled in the spatial domain. This method has been developed from its original applications in astronomy, and is now routinely used in the JET thermonuclear fusion experiment to obtain the decomposition of a spectrum of high-frequency (~10-500kHz range) magnetic instabilities with a sub-ms time resolution, allowing the real-time tracking of its individual components as the plasma background evolves. This work opens a path towards developing real-time control tools for electro-magnetic instabilities in future fusion devices aimed at achieving a net energy gain. More generally, the speed and accuracy of this algorithm is recommended for instances of physics measurements and control engineering where an unsupervised, real-time decomposition of a degenerate signal is required from a small number of data.

The JET Alfvén Eigenmode Local Manager for the real-time detection and tracking of a frequency-degenerate spectrum of MHD instabilities

“…This has prompted the developments and the applications of various methods 4 to the analysis of MHD data in thermonuclear fusion plasmas, such as the Singular Value (SVD) [18,19] decomposition of unevenly sampled data using a very small number of measurement points. As some of the mathematical background of this method has been presented elsewhere [11,12,25,26], here we only briefly review its theoretical foundations, with a more compete overview given in Appendix-A to facilitate the reading of this contribution.In the standard tokamak coordinate system (toroidal angle , poloidal angle θ), and taking explicitly into account the usual 2D boundary conditions along the longitudinal (toroidal) axis and on the plane perpendicular to it (the poloidal direction), magnetic perturbations can be represented by functions involving toroidal (n) and poloidal (m) harmonics. Considering now the usual case of a perturbation with a specific toroidal mode number n, this can be written as ( , ) i t in im mn m n e e A e…”

“…(3), and the SparSpec code uses one based on an iterative Block Coordinate Descent [11,12] algorithm. A version of this algorithm has recently been adapted to perform the required mode decomposition analysis using the rather modest computational resources available to process real-time JET data [26].3) The active MHD diagnostic system in use at JET and the Alfvén Eigenmode Local Manager.The MHD spectroscopy is a diagnostic technique that uses waves that are naturally supported by the plasma to measure the parameters that determine the dispersion relation, absorption and propagation, damping and growth of these waves [5,27]. One example of such waves is Alfvén Eigenmodes (AEs):…”

“…This has prompted the developments and the applications of various methods 4 to the analysis of MHD data in thermonuclear fusion plasmas, such as the Singular Value (SVD) [18,19] decomposition of unevenly sampled data using a very small number of measurement points. As some of the mathematical background of this method has been presented elsewhere [11,12,25,26], here we only briefly review its theoretical foundations, with a more compete overview given in Appendix-A to facilitate the reading of this contribution.…”

“…(3), and the SparSpec code uses one based on an iterative Block Coordinate Descent [11,12] algorithm. A version of this algorithm has recently been adapted to perform the required mode decomposition analysis using the rather modest computational resources available to process real-time JET data [26].…”

“…Conversely, a method based on the Sparse Representation of Signals, as implemented in the SparSpec code (freeware available at: http://www.ast.obs-mip.fr/Softwares) has been demonstrated to perform efficiently and correctly the task of a post-pulse [25] and real-time [26] blind and unsupervised signal decomposition of unevenly sampled data using a very small number of measurement points. As some of the mathematical background of this method has been presented elsewhere [11,12,25,26], here we only briefly review its theoretical foundations, with a more compete overview given in Appendix-A to facilitate the reading of this contribution.…”