Analysis of Fast Rotating Mode Signals as Precursors of Locked Modes on the DIII-D Tokamak

ASDEX Upgrade Seminar

  • Datum: 07.10.2019
  • Uhrzeit: 12:20 - 13:30
  • Vortragende(r): Michael Bergmann
  • Ort: Garching
  • Raum: Seminarraum L6, 2.Stock Süd
  • Gastgeber: IPP
This work investigates the magnetohydrodynamic (MHD) instabilities in magnetically confined fusion plasmas known as tearing modes. A tearing mode is often found to be initially rotating at high frequencies with regard to the laboratory frame and then to slow down to a stationary state ("locking"), which predates an abrupt loss of the plasma's energy and confinement, therefore leading to a so-called "disruption".The work focuses on plasma discharges relevant for modeling future tokamaks which cannot sustain multiple disruptions without being damaged. Being able to predict locked modes using precursors may help in avoiding or at least mitigating the most severe consequences. Even when rotating modes exist, they are not easily identifiable as a direct precursor of the locking; their evolution in time can vary strongly until they finally lock. We present an analysis of signals coming from a specific combination of magnetic pickup coils, capable of detecting fast-rotating modes (FRMs) while discriminating between different spatial periodicities. The analysis focuses on a special class of modes, these being likely precursors of m/n=2/1 locked modes at DIII-D.
The work is split into two parts. In the first we use amplitude and frequency of the FRM signals to evaluate physics models of mode locking on discharges relevant to the planned ITER reactor. The models focus on viscous and resistive wall forces interacting with the FRM and ultimately stopping them. We find the model to work well for discharges with high torque, while the low torque discharges lock directly without having a visible slowing-down phase. In the second part, we use the FRM signals in Machine Learning-based models, exploring their correlation features with a non-rotating radial magnetic field signal, a proxy for non-rotating modes. In addition to the FRM signals we include 12 other plasma parameters, e.g. the normalized internal inductance and edge pressure profile information, for ~250 discharges. Different Machine Learning algorithms are implemented, using both regression and classification schemes. The algorithms give the FRM signals the highest importance in predicting locked modes and are able to predict locking even in the low torque cases. However, the model is less successful in predicting discharges without a FRM signal.
Zur Redakteursansicht