Genetic algorithm optimized frequency‐domain convolutional blind source separation for multiple leakage locations in water supply pipeline

Abstract

In the realm of using acoustic methods for locating leakages in water supply pipelines, existing research predominantly focuses on single leak localization, with limited exploration into the challenges posed by multiple leak scenarios. To address this gap, a genetic algorithm‐optimized frequency‐domain convolutional blind source separation algorithm is proposed for the precise localization of multiple leaks. This algorithm effectively separates mixed leak sources and accurately estimates the delays of source propagation. Signal simulations confirm the algorithm's effectiveness, revealing that the distribution of leak positions, signal‐to‐noise ratio, and frequency characteristics of the leakage source all influence the algorithm's performance. Comparative analysis demonstrates the algorithm's capability to eliminate signal interactions, facilitating the localization of multiple leaks. The algorithm's efficacy is further validated through extensive full‐scale experiments, underscoring its potential as a novel and practical solution to the complex challenge of multiple leakage localization in water supply pipelines.