Feature extraction for acoustic leakage detection in water pipelines

Leakage detection (LD) in water pipelines is crucial for reducing water wastage. Acoustic methods for pipeline monitoring are gaining increasing popularity. However, challenges like noise, reverberation, and time-varying factors in pipelines hinder feature extraction. To ameliorate this problem, this paper introduces a feature representation method named EF_Mel spectrogram and proposes a multi-dimensional fuzzy dispersion entropy (MDFDE) for feature extraction. The pipeline acoustic signal is transformed and projected to generate the EF_Mel spectrogram. Subsequently, the features of the EF_Mel spectrogram are extracted by MDFDE. Verification of the proposed approach's effectiveness is conducted using numerical simulation and pipeline experimental bench. The results demonstrate that the proposed feature extraction is more robust in signal length, time delay, and noise rejection, achieving accuracies of 95.62 % and 96.30 % for small and large leakages, respectively, with a false negative rate (FNR) of 0 %. This paper offers a novel insight into signal feature extraction for pipeline LD.