Study on an improved acoustic leak detection method for water distribution systems

In this paper, a new method is proposed to detect leaks in the presence of the non-leak noise inside or outside a pipeline. Due to the ability to analyze the coherent structure of time series, the autocorrelation function is used to describe the self-similarity feature of the leak signal. The values of the autocorrelation function for the lag larger than the correlation length of the signal, not the signal itself or the entire autocorrelation function, are used to extract or evaluate the self-similarity degree of the signal by the approximate entropy. Based on feature extraction, a new detection function related to autocorrelation functions of the acquired signals is built to detect leak. Then a neural-network approach is utilised as a classifier to discriminate the leak signals from the non-leak signals inside and outside pipes. The proposed method has been employed to detect leak in the presence of the non-leak noises inside and outside pipes, and achieved a 93.8% and 86.3% correct detection rate, respectively.