Flow Feature Extraction And Image Reconstruction For Electrical Capacitance Tomography System

ECT (Electrical Capacitance Tomography) is a kind of tomography technology. In ECT, capacitance values of the pair of electrodes around the object are measured, according to which permittivity distribution of the object can be computed, and spatial distribution inside of the object can be obtained. Because of some advantages such as non-intrusion, widely usage, safe, fast response, simple structure and so on, ECT is mainly used for detection of multiphase-flow in industrial pipeline. The successful application of ECT depends on the speed and precise of image reconstruction. In this paper, 12-electrode ECT system is studied, and the research on the key technical issues of flow feature extraction and image reconstruction are carried on, and the main contents are as follows:ECT system is discussed in the composition detail and working principle. The mathematical model of ECT system is built, and according to the mathematical model of measured field, solution of forward problem of ECT is analyzed. FEM (Finite Element Method) is adopted to establish the relationship of capacitance values, sensitivity distribution and gray value of image, and the principle of image reconstruction is discussed, which are the theoretical basis of flow feature extraction and image reconstruction.According to the effect of structural parameters for the sensor on system performance, the mathematical model of sensor field sensitive is established. From the perspective of soft field, media distribution, sensitive field uniformity and structure parameters of the sensor are analyzed, and optimization function of sensor design is determined. To obtain the best structural parameters of sensor, response surface methodology is adopted to fit the implicit function. Optimization design of sensor is finished by solving the RSM function to obtain a set of sensor optimization parameters.For flow feature extraction and identification problems, different flow characteristics of the capacitance value of the distribution are analyzed. According to the distribution of different flow, 11 feature parameters are defined, which include feature information of different flow regime. Based on the analysis, feature extraction and RBF neural network are brought forward to solve flow pattern identification problems. In this method 11 feature parameters are input into RBF neural network to train, and efficiency of identification is improved. Experiment of simulation results show that for the typical stratified flow, full of water, annular flow and uniform flow, the average recognition rate that using RBF neural network is more 10% than in BP neural network, and the performance is better than inputting raw capacitance values to neural network.For ECT forward problem solution and identification problem, the capacitance values in ECT forward problem are as a priori knowledge to establish multivariable relationship of test data, and select principle component analysis to reduce the dimensions of multiple variables, and the comprehensive factors are not correlated, and raw variables are most included. After dimensionality reduction, SVM (support vector machine) is adopted to train the training sets, classification models and classification decision functions obtained are used to classify 6 flow pattern: laminar flow, core flow, droplet flow, annular flow, full of oil, and full of water. The impact of different parameters of SVM on experiment results is discussed. GA (genetic algorithm) and PSO (Particle Swarm Optimization) are applied to optimize the parameters, and optimized SVM model is better for high classification accuracy.With consideration of "soft field" and ill-posed problems, an ITR (Improved Trust Region) ECT image reconstruction algorithm is brought forward. In this algorithm, the procedure of steps to solve nonlinear ill-posed problems is presented, and Hesse matrix is corrected, which is generated in the process of iteration according to BFGS formula. At last convergence of the algorithm is analyzed and proved for ECT image reconstruction, which supplies an effective method for ECT image reconstruction.