Spectral statistical analysis of low frequency coefficients from diagnostic signals depicting MHD disruptions

Abstract

Aditya Tokamak is a fusion reactor for obtaining nuclear fusion energy from high temperature, ionized gas called plasma. The magnetic field is used to confine plasma in the shape of torus. A disruption is a violent event that terminates magnetically confined plasma. In a disruption, the temperature drops drastically and heat particles are released from confinement on a short timescale and dumped on the vessel wall, causing damage in proportion to the stored energy. The loss of confinement is associated with the production of runaway electrons, which may also produce damage. In order to mitigate disruption, it is necessary for early prediction of disruption. The signals like Halpha, Hard X ray, Plasma current, Mirnov coil signal, Vloop, Soft X-ray should be analyse for the detection of plasma disruption. From the signal if there is any peak in the Halpha, Hard X ray and the non negative value appears in Vloop signals before the decayed of the plasma current indicates the disruption. There are hard and soft disruptions. Since the hard disruption is dangerous, methods like Artificial Neural Network (ANN) using mmultilayer perception and the Fast Fourier Transform (FFT) where already used to find the disruption in terms of frequency component signals. Since Aditya Tokamak signals are non stationary, the above mentioned methods for stationary signal, so those methods are not providing correct and satisfactory results. In this work, different wavelet transforms like Daubechies, Discrete Meyer, Symlets and Biorthogonal were applied on the signal and the corresponding approximation and detailed coefficients were obtained from all the signals in order to obtain the disruption. On an average, of thirty signals are used from different shots for the analysis. Out of the above mentioned wavelet transforms, Discrete meyer and Biorthogonal wavelet are providing the better results than others in terms of the statistical parameters such as mean, skewness and kurtosis. Mean is minimum and the skewness and kurtosis are maximum in the disruption signal, which is confirmed with its time domain analysis. Discrete meyer and the Biorthogonal wavelet transforms provide the spectral information in contrast to frequency domain approaches like FFT. It provides early information about the hard disruption. Wavelet transform is better than FFT and ANN. Since the range of parameters responsible for disruption is not fixed by this method the analysis can be strengthened with Framelet transform.