Extraction of Sound Signals from Power Cable Impulse Discharge

Abstract

In the fault detection of buried power cables, random noise in the shock discharge sound signal is difficult to remove by existing filtering methods. A fifth-order convergent independent component analysis (ICA) method based on empirical mode decomposition (EMD) is proposed to extract the impact discharge sound signal. The fifth-order convergence ICA is adopted, so that the eigenmode components decomposed by the EMD and the remaining signals are independent of each other. Using the strong correlation between the frequency spectrum of the discharge sound signal, the eigenmode component with the largest correlation between the frequency spectrum and the high signal-to-noise ratio shock discharge sound signal spectrum is automatically extracted. Finally, the shock discharge sound signal of unknown failure point is obtained. This method has the advantages of less constraints, small dependencies, and fast convergence. The simulation and experimental results further show that the discharge sound signal in the mixed signal can be effectively extracted. Introduction Buried cables are prone to failure, but the point of failure is not easily detected. If the fault cannot be dealt with in time, it will cause serious consequences. Therefore, it is of great significance to find a fast and accurate method for locating cable faults. At present, many methods are used in cable fault location, including sonic detection method, magnetic sound synchronization method, audio current induction method, etc. [1]. In these methods, the cable discharge sound signal needs to be detected and extracted. The cable discharge sound signal at the fault point of the cable is extremely susceptible to the influence of ambient noise, which makes it difficult to process the signal. The noise in the cable discharge sound signal is non-stationary random noise. Traditional filtering methods, such as the classic spectral analysis of fast Fourier transform, parametric autoregressive moving average spectral analysis [2], Kalman filter [3], etc., are difficult to remove this noise. The empirical mode decomposition method can decompose mixed signals in engineering practice [4]. However, the empirical mode decomposition method will cause aliasing of the eigenmode components of the signal, which is not conducive to the extraction of the signal. Aiming at the problems of signal decomposition by empirical mode decomposition, scholars at home and abroad have proposed many methods [5], but the processing effect is not ideal. A fifth-order convergent ICA method based on EMD is proposed to extract the impact discharge sound signal, and the effectiveness of the method is proved by simulation and field experiments. EMD-based Fifth-order Convergence ICA ICA, which is applied to the extraction of useful sound signals, has achieved good results [6]. However, ICA can be solved only when the number of observation channels is not less than the number of source signals in the mixed signal. In practical applications, only one observation signal can be obtained. As a result, the use of ICA has been reduced. EMD can decompose a single signal. During the decomposition process, EMD cannot completely guarantee that the decomposition components are independent or orthogonal to each other, and the modal components obtained by the decomposition have modal aliasing [7]. Through the analysis of the two methods, ICA and EMD are combined. This method can not only decompose a single signal, but also make the decomposition results independent of each other. The initial separation matrix in ICA is randomly selected, which will cause a large difference in the number of iterations and even non-convergence. Therefore, the relaxation factor  is used to overcome the dependence on the initial value and increase the convergence range. The objective function of ICA algorithm based on the largest negative entropy is Eq. 1. ( ) { ( )} T k k k F W E Xg W X W    (1) In the formula, k W is the separation matrix, k is the number of iterations, E is the mean operation, X is the observation data, g is a non-linear function, and { ( )} T T k k E W Xg W X   . is a relaxation factor, which has a characteristic of continuously decreasing the objective function value.  is represented by Eq. 2, that is, Eq. 3. 1 1 { ( )} { ( )} T T k k k k E Xg W X W E Xg W X W        (2)