Improvement of pipeline leak detection method: Integration of spectral entropy and sample entropy for better description of complexity features

Abstract

Entropy measures signal complexity and can aid in pipeline leak detection, but most studies rely on time-domain entropy alone, overlooking spectral complexity. In the vicinity of pipeline systems, there are often complex machines operating continuously. During the experiment, noise from an air compressor was often mistaken for a leak signal due to overlapping time-domain entropy characteristics. To address this, we analyze the failure mechanisms of time-domain entropy features in such scenarios by constructing representative signals and propose a pipeline leak detection method based on Spectral Entropy-Sample Entropy (SE-SampEn). This method better captures the complexity of leak signals, improving detection accuracy under diverse noise interferences. Firstly, the primary components of the signal are extracted through variational mode decomposition (VMD), enabling the removal of background noise during the SE calculation process. Subsequently, an applicability analysis of selected signal features is conducted, and feature vectors are constructed and input into various classifiers to achieve effective leakage detection in the water supply pipeline. The results demonstrate that the accuracy reached 97% when both SE and SampEn were utilized, but dropped to 84% and 87% when SE and SampEn were omitted, respectively. This indicates the necessity of using both SE and SampEn to fully describe the complexity feature of the leakage signal in complex noise environments.