Research On Water Supply Pipeline Leak Detection Expert System Based On Analysis Of Leakage Sound Characteristics

Water supply network is one of the important infrastructures in urban and rural areas,which is mainly composed of various pipelines buried underground.These pipelines will inevitably leak in the long run.When the water supply pipeline leaks,the leakage point will emit a specific frequency sound under the action of pressure.At present,the commonly used leak detection method is the manual leak detection method.Leak detection workers use leak detection rod to find the leak point in the pipeline.Artificial hearing leakage is simple and convenient,but the diagnosis results are closely related to the experience of hearing leakage workers,and there are problems of low efficiency and low accuracy in leak detection.Therefore,how to combine traditional artificial hearing leakage with deep learning neural network and other artificial intelligence methods to design and develop the water supply pipeline leak detection and diagnosis expert system,so as to improve the efficiency and accuracy of water supply pipeline leak detection has become a research focus in the field of pipeline leak detection.This paper combines traditional leak listening technology,expert experience and artificial intelligence technology,constructs a water supply pipe leak sound dataset,establishes a deep learning neural network-based pipe leak sound recognition model,and designs and develops an expert system for pipeline assisted leak detection based on a cloud server platform.The main work and research contents of this thesis are as follows:(1)A pipeline leakage sound data collection device was designed and developed,and feature extraction and analysis of leakage sounds were carried out.A complete set of water leakage sound collection device was designed using a leak listening rod as the front-end listening device,and a software algorithm was used for data noise reduction processing to achieve a realistic and effective recording of water leakage sounds.Experimental results show that the logarithmic power spectrum and Log Mel spectrum of water leakage sound can be effectively identified by extracting and visualizing these two features.(2)A pipe leakage sound dataset was established,and the validity of the dataset was verified and analyzed.Firstly,4.7 h of pipe leakage sound data were collected in the field using the leakage sound collection device designed in this paper,covering 6types of pipes,6 pipe diameters and 16 types of leakage points.Seven event tones were then selected as interference data and combined with the collected data to create a raw data set of 9.73 h in length.Further,to address the data imbalance in the dataset,an offline data enhancement method was proposed to add background ambient noise,fusing the four ambient noise data with the original data,resulting in an augmented dataset of 69.27 h,effectively increasing the complexity and quantity of the data.Finally,a support vector machine(SVM)-based classification method was given using the Meier cepstrum coefficient(MFCC)as input,and classification experiments were conducted on the dataset.The experimental results show that the SVM model has a recognition accuracy of up to 92.8% and a detection rate of 94.8% for water leakage sounds,and the SVM as a baseline model is very effective in distinguishing water leakage sounds.(3)A deep neural network-based pipe leakage sound recognition model was established to verify and analyze the effectiveness of the model.On the produced dataset,four sets of experiments were conducted using the shallow network Mushtaq Net and the deep network VGG11 as the backbone networks,and the log power spectrum and Log Mel spectrum as the inputs respectively.The two convolutional neural networks were improved for the shape characteristics of the inputs,and the improved networks reduced the number of parameters and avoided overfitting.The experimental results show that:(1)the Log Mel spectrum features are more distinguishable,and the experimental results of the same model using the Log Mel spectrum are better than those using the logarithmic power spectrum;(2)in the experiments of this paper,the experimental results of the VGG11 type network using the Log Mel spectrum as input are the best,and the recognition accuracy for water leakage sounds is as high as 97.77% and the detection rate is 98.17%.(4)An expert system for auxiliary leak detection in water supply pipes was designed and developed and experimentally validated.In this paper,the resulting model was deployed in the cloud,and the client application software was designed to realize a cloud service expert system with front-end data collection,cloud model algorithm calculation,and client display of diagnostic results.The functions of the system and the effectiveness of the model algorithms were verified in a real working environment.