Inspection of Disruptive Behaviours at JET using Generative Topographic Maps
Giuseppe A. Rattá Gutiérrez, Andrea Murari, Jesús Vega, Guido Vagliasindi
Tokamaks are the most promising configuration of magnetic confinement fusion devices. Actually, the biggest and most important machine of this kind is the Joint European Torus (JET), located in Culham (UK). Besides the advantages that Tokamaks have demonstrated to provide, a dangerous physical phenomenon remains unavoidable and risks its integrity. When this abnormal behaviour, called disruption, occurs, the plasma (a heated and ionized gas) confinement is suddenly lost and in an amount of time of tens of milliseconds its energy content is transferred to the first wall. In addition, eddy currents are induced over the surrounding structures causing high forces over them.
To measure the physical quantities of interest inside the vacuum vessel, advanced sensors systems are attached to the device. Those diagnostics transforms the acquired quantities into electrical signals. After every experiment, the resulting data are mostly temporal evolution waveforms but also images, contour plots, profiles and scatter graphs. The information provided by the diagnostics can be utilized to detect unusual instabilities or disruptions precursors to notice in advance the phenomenon and consequently to apply control or mitigation actions to reduce the possible damages. Due to the complex nonlinear interaction of the involved variables that lead the plasma to its abrupt end, nowadays has been impossible to develop a complete an unfaultable theoretical model to prevent their occurrence. In this article two important aspects that can facilitate the better understanding of the phenomenon are presented. First, the selection of the physical measurements [1, 2, 3] and their main characteristics related to disruptions is reviewed [4]. Second, the application of Generative Topographic Maps (GTM) [5] to visualize and compare the evolution of disruptive and non disruptive experiments is detailed. This unsupervised method can be considered as visual proof of the evolution of the phenomenon.