Imaging Algorithm Of Gas Leakage Source Based On Deep Learning

The leakage problem in gas transportation and storage is closely related to production,life and safety,which has important economic value and practical significance.The localization and imaging of gas leakage source is also an important research direction in the field of industrial nondestructive testing.Acoustic detection method has the advantages of wide range of application and no gas selectivity,which has been widely studied.Traditional acoustic methods are mainly divided into three categories:high-resolution spectrum estimation methods,time delay estimation methods,and beamforming methods.However,these algorithms are susceptible to environmental noise,array shape,array aperture and other factors,resulting in a serious decline in imaging accuracy.In order to optimize the problem,an imaging algorithm of gas leakage source based on deep learning is proposed by using virtual array.Virtual array can break through the problems of fixed array and limited array aperture,reduce the cost and complexity of the system,and ensure the accuracy of imaging detection.Compared with traditional algorithms,this method can achieve high-precision imaging under the conditions of fewer array elements,more classification grids,lower signal-to-noise ratio and longer collection distance,which has higher imaging accuracy and stronger robustness.Firstly,the leakage signal model and virtual array signal model are analyzed,and the time interval caused by scanning is eliminated by cross power spectrum method.Secondly,the theoretical basis of deep learning is introduced,and the residual neural network is built.The effects of spatial sampling rate,number of classification grids,signal-to-noise ratio and collection distance on the imaging performance are compared by simulations.Finally,the experimental platform is built and the gas leakage experiments are carried out.The gas leakage data are collected with a 20×20 and a 30×30 array at a distance of 0.68 m and 0.3 m,respectively,and the phase information are used to image the leakage source.The results of 15 repeated experiments show that the accuracy of the proposed algorithm is 73.3%,80.0%,86.7%and 93.3%,respectively,which proves the high imaging accuracy and strong robustness of the proposed algorithm.The imaging algorithm of gas leakage source based on deep learning is simple and intuitive,which is helpful for operators to identify and take measures.