Spectral-Based Methods for Pipeline Leakage Localization

Abstract

In this paper, the pipeline leak localization problem using transient data is investigated. Signal processing techniques that proved successful in wireless communications and acoustics are adapted and tested for leak identification. More specifically, Bartlett’s beamforming (BF) (also known as conventional BF, matched field, or phased array), Capon’s BF (also known as the minimum variance distortionless response filter), Lagunas’ BF, and multiple signal classification (MUSIC) methods are used. The localization is realized by a one-dimensional search for the leak location along the pipe, where one-dimensional search means that the wave model used includes one leak only. The one-dimensional search is advantageous in that it involves low computational cost. The performance of the different techniques in the cases of a single leak and multiple leaks is discussed. In the single-leak case, the proposed spectral methods accurately localize the leak even for a high level of noise. For the multiple-leak case, the proposed spectral methods are able to localize all leaks provided that the leak-to-leak distance is of the same order or larger than half the shortest probing wavelength. However, the localization deteriorates when the leaks are too close together because a model with a single leak is being used to identify multiple leaks. Although not accurate, the application of the one-dimensional search to multiple leaks is still valuable because it provides a fast initial estimate of the leak locations, which serves as prior information for more precise but computationally expensive multidimensional search techniques.