Pipeline leak detection through implementation of empirical mode decomposition and cluster analysis

Pipelines are crucial for the effective and safe transmission of natural resources across extensive distances. Nevertheless, they are highly vulnerable to the influence of environmental factors. Therefore, regular inspections are crucial. This paper introduces a novel methodology that combines Empirical Mode Decomposition (EMD) with K-means clustering to improve noise reduction and precisely locate pipeline leaks in real-world scenarios. Acoustic data was gathered from three distinct places at varying air pressures near simulated pipe leakage in an open environment. The acquired signals were significantly influenced by environmental noise. To resolve this, EMD was utilized to decompose the signals into intrinsic mode functions (IMFs). The selection of final IMFs for subsequent analysis was based on the optimal signal-to-noise ratio. Subsequently, silhouette analysis was utilized to optimize the most suitable number of clusters for each selected IMF. Then, K-means clustering was implemented for the chosen IMFs. The approach effectively detected leakages in the time domain for each scenario: 0.837 s for case-1, 1.26 s for case-2, and 2.35 s for case-3. A thorough statistical study confirmed the clustering outcomes, showcasing accurate and dependable leakage detection, characterized by low variance, skewness, and kurtosis values that support the method's reliability. The results demonstrate that the K-means clustering approach is extremely dependable and accurate in detecting leakages. This significantly improves the reliability of pipeline inspections and offers a practical approach to identifying and managing breaches at an early stage.