Early Identification of Disruption Paths for Prevention and Avoidance

Disruption prevention in high performance, high current, long duration discharges requires a substantial evolution of the schemes applied in most of the present tokamaks. An efficient prevention scheme requires the early identification of the nature of the off-normal behaviour and the automatic selection of the appropriated countermeasure, either avoidance or mitigation. For the purpose of the avoidance, on which this paper is focused, the disruption can be seen as the result of the interplay of the physical events and of the control system responses to them and to the technical failures. The building blocks of such description should include the integration of several sets of plasma scalar data, plasma profile data, magnetohydrodynamics (MHD) indicators and engineering data. Previous work has shown the potential of the Generative Topographic Mapping (GTM) algorithm for identification and discrimination of the disruptive operational space in tokamak devices. In the paper it is shown that the magnetic fluctuations associated with rotating MHD modes can be characterized using a set of observables derived from the Singular Value Decomposition applied to the data collected by an array of pickup coils. They provide an input to the GTM analysis such that a clustering separating disruptive and non-disruptive timeslices can be found. A selection criterion is derived from this analysis such that a warning time of the order of seconds about the incoming disruption can be obtained.