Research On Gas Leakage Detection Method Based On Acoustic Signal

Various gases are widely used in industrial production.Some gases are flammable and explosive,and some are toxic.If leakage occurs,it may cause major safety accidents.Therefore,it is necessary to carry out gas leakage detection research.In this paper,gas leakage detection is taken as the research object.Based on signal analysis and processing technology,the problem of gas leakage detection is studied by using acoustic characteristics combined with machine learning algorithm and deep neural network.The main research contents are as follows:(1)The acoustic signal acquisition device is designed to analyze the acoustic signal of gas leakage.According to the specific requirements of the application scene acoustic signal acquisition,select the appropriate hardware model,design and implement a small four-element microphone array acquisition device.The small array is used to collect the acoustic signal of gas leakage,and the influence of gas type,small array acquisition distance and gas leakage pressure on the acoustic signal of gas leakage is analyzed.The experimental results show that the gas type has little effect on the acoustic signal of gas leakage,while the exhaust pressure has a great influence on the acoustic signal characteristics.(2)The gas leakage characteristics are extracted to study the machine learning model of leakage signal detection.Firstly,according to the analysis results of the gas leakage signal,the Spectrum Centroid,Spectrum Bandwidth,Spectrum Contrast,Spectrum Flatness,Short-Time Zero Crossing Rate and Mel-Frequency Cepstral Coefficients of the gas leakage acoustic signal are selected to construct the gas leakage feature vector.machine learning method is used to establish integrated learning and linear learning models;finally,statistical learning evaluation indexes such as recognition accuracy,recall rate,false positive rate and area under the receiver working characteristic curve of different models are compared.The experimental results show that Ada Boost model is better than other models on the whole.(3)The deep learning neural network model of gas leakage detection based on short-time Fourier transform is studied.Firstly,different background and leakage acoustic signals are collected by simulating the gas leakage environment.The collected digital acoustic signals are transformed into two-dimensional images containing timefrequency information by short-time Fourier transform,and the experimental data set is constructed.In view of the large calculation amount of the classical VGG neural network model,the full link layer of the network is changed to the global average pooling layer,which reduces the calculation amount and ensures the calculation accuracy.Select the appropriate optimization algorithm and loss function,adjust the training batch size and learning rate of the model,and propose VGG-GAP model;finally,the gas leakage experiment proves that the VGGG-GAP model has good performance in the application of gas leakage detection..In this paper,the traditional acoustic characteristics and statistical learning methods are combined to study the gas leakage detection,and good results are obtained.It has certain guiding significance for the industrial detection of gas leakage.