Improving Pipeline Leak Detection Accuracy by Constructing a Multistate Leakage Sounds Training Set

Acoustic-based pipeline leakage detection technology is a current research hotspot, where leakages can be identified by detecting acoustic waves generated by turbulent pulsations and cavitation. However, the turbulent structure at the leakage site varies with the shape of the leakage orifice, which may lead to variations in the features of the leakage sound signal, thus affecting detection performance. To address this issue, this paper proposes a leakage detection method that considers multistate acoustic signals. Leakage signals under different orifice shapes were experimentally collected and analyzed for their time-frequency features; then, feature vectors of multistate leakage signals were integrated to construct the training set. Finally, a suitable kernel function was selected, to achieve a highly accurate leakage detection based on a support vector machine (SVM) classifier. The experimental results show that compared with a single-state leakage signal training set, the overall accuracy of the multistate signal training set was significantly improved, reaching a maximum of 100%. This study validates the importance of collecting multistate leakage signals during the construction of training sets for leakage detection tasks and provides valuable guidance for data acquisition and feature selection in leakage detection applications.