Research And Optimization Of Algorithm For Capacitively Coupled Electrical Resistance Tomography Of Two-phase Flow

In industrial production,it is very common for two or more media to mix in pipes.However,accurately measuring the related parameters is challenging due to the complex mixing situation in pipes and the variable flow state of media.Additionally,the corrosiveness of some media affects measurement accuracy.Capacitive coupling resistance tomography imaging technology is a new type of tomography imaging technology that combines capacitive coupling non-contact conductivity detection technology and traditional resistance tomography imaging technology.It has garnered significant attention due to its advantages of being non-contact,having a simple structure,being cost-effective,and demonstrating good robustness.However,traditional imaging algorithms are highly sensitive to errors due to the nonlinearity,ill-posedness,and uneven distribution of sensitive fields in inverse problems.As a result,high-quality reconstructed images cannot be obtained using these algorithms.In recent years,deep learning has shown great potential in image processing,leading to the emergence of applying neural network algorithms as a new research direction for solving inverse problems.The main research contents of this paper are as follows:The composition and operational principles of the Capacitive-Coupled Resistivity Tomography(CCERT)imaging system were thoroughly analyzed.A 12-electrode imaging model was constructed using the COMSOL Multiphysics software,which is a multi-physics simulation tool.The theoretical analysis was conducted for both the forward problem and the inverse problem.The sensitivity field distribution of the CCERT sensor was investigated,and the findings revealed that the CCERT sensor exhibits no negative sensitivity region and the sensitivity field distribution is non-uniform.The sensitivity is higher in the vicinity of the electrodes.Several traditional reconstruction algorithms,namely the LBP algorithm,Tikhonov algorithm,Landweber algorithm,ART algorithm,and TSVD algorithm,were employed to reconstruct images from simulated data of a two-phase flow.The reconstructed results were thoroughly analyzed and compared using metrics such as relative image error and correlation coefficient.The outcomes demonstrated that the iterative algorithms yielded superior image reconstruction results.To obtain a data set for neural network algorithm-based image reconstruction,the CCERT model was utilized to capture the data of two-phase flow with different positions,sizes,and quantities of target objects.The data set comprised two components: resistance measurements and conductivity distributions.BP and RBF neural network algorithms were employed to reconstruct images of three types of two-phase flow patterns.A comparative analysis was carried out against traditional algorithms.The results showed that the neural network algorithms effectively reduced image errors,improved correlation coefficients,and resulted in enhanced image quality compared to traditional imaging algorithms.In order to address the issues of artifacts,unclear edges,and further enhance image accuracy in BP and RBF neural networks,a novel image reconstruction algorithm based on multi-classification convolutional neural networks(CNN)was proposed.The CNN algorithm was employed to reconstruct images of the three types of two-phase flow patterns and was compared against BP and RBF algorithms.The results demonstrated that the CNN algorithm accurately depicted the distribution of the medium within the pipeline,and it significantly outperformed other algorithms in terms of image error and correlation coefficient analysis.