A model-based health indicator for leak detection in gas pipeline systems

Abstract

Leakage in gas pipelines is becoming a significant issue and has attracted much attention in recent years. This paper is concerned with the development of a robust health indicator for identifying the leakage in gas pipeline systems. A spectral exponent indicator is proposed based on a theoretical leak noise spectrum model. Measurements of the leak acoustic signals are also presented from a pipe rig with air under pressure. Then, a feature selection technique is employed to select properly desired features. Three data-driven approaches, artificial neural networks (ANNs), support vector machine (SVM), and random forest (RF) are trained with the most discriminative features. The proposed methodology showed to achieve 99.4%, 99.6% and 99.4% accuracies for ANN, SVM and RF respectively. Furthermore, the proposed indicator showed to be robust under different conditions illustrating its ability for applications in the field.