Enhanced leak detection in water distribution systems using GFCC-based signal processing techniques

Abstract

Leak detection in water distribution systems remains a significant challenge, especially for small-scale leaks masked by noise. This study proposes an advanced signal processing approach using Gammatone Frequency Cepstral Coefficients (GFCC) combined with a squeezing technique to enhance the detection of weak leak signals. Experiments were conducted using a 152-m MDPE pipeline under two water pressure levels (1 bar and 2 bar) and five leak sizes (1 mm–5 mm). The proposed method detected leak signals at low signal-to-noise ratio (SNR) levels as low as 95 dB, where conventional methods typically fail to do so. Specifically, the squeezing technique effectively amplified transient leak signals, enabling accurate identification of leak-induced spikes within high-noise environments. The analysis revealed a significant improvement in detection, with leak signals becoming distinguishable even under severe masking conditions. The proposed GFCC-based technique achieved consistent detection across 30 repeated datasets, validating its robustness and repeatability. The novelty of this study lies in integrating GFCC and squeezing, which were initially used in speech signal processing, into transient-based leak detection in pipelines, an area that has not been widely explored. This hybrid technique presents a promising solution for reliable and high-sensitivity leak detection in modern water distribution networks.