Evaluating the impact of spectral estimators on frequency domain feature classification applications for pipe leakage detection

Abstract

In this paper, the spectral feature selection process for leak detection on pipelines carrying fluids is studied, with focus on the impact of the adopted spectral analysis and estimation tools. The leak detection is treated as a binary classification task, utilizing a series of frequency domain features, which are estimated from the spectral data of captured, experimental, leak signals. The performance and accuracy of the detection process is comparatively tested, when utilizing various spectral estimators as the power estimation block, which is present in the core of every spectral descriptor’s algorithm. The well-known Periodogram, Welch and Capon estimators are considered. The illustration of the analysis results is facilitated by the use of popular classification performance indicators.