Water Supply Pipe Leakage Detection Based On Acoustic Signal Characteristics

Every year due to natural factors such as pipeline corrosion and geological subsidence and human factors in urban development and construction,urban water supply pipelines will inevitably leak.If pipeline leakage and leakage conditions can be discovered in time,a large amount of water resources will be saved.Therefore,water supply pipeline leak detection plays an important role in the entire system of monitoring the water supply network.Pipeline leakage detection methods are mainly based on two kinds of acoustic signals and non-acoustic signals,and mainstream methods based on non-acoustic signals include flow balance method,infrared thermal imaging method,and model method.Due to restrictions imposed by pipelines and high precision requirements,the current mainstream detection method is based on the detection of acoustic signals.Earlier adopting the listening method completely relied on the experience of the workers.The main research contents of the later researchers were: 1)The analysis of the position of the peak of the vibration spectrum spectrum of the pipeline can be used to determine whether the pipeline is leaking.However,this method works against the spectrum of the leakage signal.Similar narrowband noise cannot be discerned.2)Combine training data with signal decomposition theory and machine learning algorithms such as artificial neural networks to discriminate leakage signals.This method is highly dependent on the amount of data.This paper proposes a machine-learning-based leak identification model specifically for small sample processing.The specific content includes:(1)Analyze the main form of pipeline leakage and the excitation source that causes pipeline vibration at the leakage outlet in the event of a leak.Analyze the frequency spectrum difference in the presence or absence of leakage and the effect of pipe material on the time domain and frequency domain of the signal.(2)The use of Huffman coding for lossless fusion of pipeline signals at the data level is conducive to data transmission.The LMS adaptive cancellation algorithm is proposed to filter the background noise of the pipeline,and the signal is filtered.(3)Based on the difference in characteristics of the leakage signal and the non-leakage signal,after the empirical mode decomposition of the signal,the average power spectral density characteristic of the intrinsic modal function,the approximate entropy characteristic of the original signal,and the principal component of the representative signal are extracted.Simulation analysis is based on the leakage recognition effect of single signal features at different signal-to-noise ratios.Then combine multiple feature composition feature matrices,use the binary support vector machine to train the data feature matrix,test the trained network with known signals,and improve the accuracy by optimizing the network.Finally,the optimized network is used to detect the actual water supply pipeline discriminating leakage.