Speaker
Barbara Cannas
Description
See the full Abstract at http://ocs.ciemat.es/EPS2018ABS/pdf/P4.1017.pdf
Wavelet analysis of Mirnov coils signals for disruption prediction at JET
B. Cannas1, A. Fanni1, A. Murari2, A. Pau1, G. Sias1 and the JET contributors*
1
Electrical and Electronic Engineering Dept. - University of Cagliari, Italy
2
Consorzio RFX (CNR, ENEA, INFN, Università di Padova, Acciaierie Venete SpA), Corso
Stati Uniti 4, 35127
*See the author list of “X. Litaudon et al 2017 Nucl. Fusion 57 102001
Disruption are often preceded by MHD modes slowing down, growing and locking when the
amplitude exceeds a critical value. The fluctuations of the poloidal magnetic field recorded
by the Mirnov coils could provide useful markers related to the presence of this kind of
instabilities causing disruptions. This paper proposes a time-frequency analysis of the
Mirnov signals recorded at JET by the high resolution probes of KC1M diagnostic. The work
is a contribution toward the definition of new features characterizing disruptive behaviours to
be used as input in a multisignal disruption predictor.
Mirnov coil signals are non-stationary, thus, their spectrum changes with time. Compared to
Fourier transform, wavelet transform (WT) is a step forward in the spectral characterization
of a time series, since allows to study the temporal evolution of amplitude, frequency over
time scales comparable with the wave period. The continuous WT of a discrete time series
{x[i]} is defined as the convolution product of {x[i]} with a scaled (t → t/s) and shifted (t →
t − τ) version of a mother wavelet ψ(t). The windowing is intrinsic in the wavelet and it
depends on scale s. The smaller s, the more compressed the wavelet is, so it can catch rapidly
changing details in the signal.
The database taken into account in this work consists of 116 non disruptive and 78 disruptive
pulses. The Mirnov signals are acquired at the frequency of 2MHz and resampled at 20kHz.
Then, the wavelet coefficients are evaluated on a mobile window as in a real time application.
The results of a preliminary analysis performed on a single probe show that, for all non
disruptive pulses, several high scale (low frequency) coefficients stay smaller than an
optimized value, whereas for 86% of the analysed disruptive shots they exceed this value
more than 10ms before the disruption time. Thus, this technique seems promising to
anticipate the alarm of a multisignal predictor in case of rotating precursors to locked mode.
In the final paper, an analysis of more signals from several Mirnov coils will identify the
presence of locked modes or other instabilities in order to interpret the results. The study will
be extended integrating the information given by different wavelet coefficients.
