Information processing for leak detection on underground water supply pipelines

In the correlation-based leak location, it is supposed that the correlative components in the spatially separately collected acoustic signals merely result from a leak or leaks. This is why a false leak location will be produced when there is a non-leak acoustic source occurring outside a pipeline. To void a false leak location, it is necessary to detect whether or not a real leak exists in the pipeline beforehand. The traditional methods can detect leak only when the leak signal and non-leak signal are not acquired by the vibration sensors in the pipeline simultaneously. However, in practice, the leak signal is always blurred with the non-leak signal, and they will be picked up by the vibration sensors simultaneously. In this case, the traditional detection methods are infeasible. In this paper, a new feature extraction and leak identification method using autocorrelation analysis and approximate entropy algorithm is proposed to detect leak in the presence of non-leak acoustic sources. Due to the ability to analyze the coherent structure of time series, the autocorrelation function is used to describe the self-similarity of the signal. And the autocorrelation function values for the delay τ larger than the signal correlation length, not the signal itself or its entire autocorrelation function, is used to extract or evaluate the self-similarity degree of the signal by the approximate entropy algorithm. A neural-network approach has been developed as a classifier, which uses the identified self-similarity degrees as the network inputs. The method has been employed to identify the leak signals in the presence of some non-leak sounds, and achieved a 93.8% correct detection rate.