Research On 2D Image Reconstruction And Flow Pattern Recognition Of Electrical Capacitance Tomography

Electrical capacitance-tomography(ECT)is a nondestructive imaging technology using two-phase/multiphase flow to measure or monitor the flow process parameters by visualizing the material distribution in a pipe or closed vessel.ECT system has the advantages of non-intrusion,non-radiation and simple operation,so it has high application and research value.The main work contents are as follows:(1)An image reconstruction algorithm based on overcomplete dictionary is proposed.The optimization equation of ECT inverse problem was reconstructed according to the principle of image sparse representation,and the nonlinear mapping relationship between observation capacitance and sparse space was directly established.Samples were generated automatically through programming and K-SVD algorithm was used to learn the over-complete dictionary.Landweber,Tikhonov and Newton-Raphson algorithms were used to optimize the solution of characteristic coefficient vectors,and then the distribution of dielectric constant was obtained.The simulation results show that the proposed method can ensure the accuracy of image reconstruction.(2)An image reconstruction algorithm based on deconvolution network is proposed.The instability and discomfort of the inverse problem are the main obstacles that need to be overcome by the image reconstruction algorithm.This problem is often transformed into the problem of how to design the minimization of the regularization constraints.Two-phase flow image has semantic characteristics and the ideal representation of two-phase flow distribution in cross-section image is the pixel region with distinct boundary.The process of ECT reconstruction was regarded as a semantic segmentation problem.Two kinds of deconvolution network models were designed according to the nearest neighbor interpolation up sampling and transposed convolution respectively.Based on the nonlinear fitting of the network,the 1-D capacitor sequence was mapped to 2-D gray image.Through the simulation modeling of ECT system,it is verified that the proposed method can image more quickly and with higher accuracy than the classical ECT image reconstruction algorithm.(3)A two-phase flow pattern recognition method based on Broad Learning System(BLS)is proposed.The fusion method of artificial features and BLS network is proposed,and three kinds of artificial features are extracted,which are the number of regional blocks,circularity of regional blocks and the center of gravity of the region block.Single artificial feature and BLS network fusion methods and multiple artificial features and BLS network fusion methods were designed to explore the influence of different artificial features on the convection recognition rate.The contribution of each feature is evaluated by simulation experiments and the effectiveness of the proposed method in predicting two-phase flow patterns is verified.