Research On Image Reconstruction And Flow Pattern Identification Method For Electrical Capacitance Tomography

In industrial applications,operators often need to understand the flow of material inside closed pipes or machines in order to take appropriate measurement,analysis,and control methods for industrial processes.The flow of material within a non-visible environment is generally a mixed flow,which is difficult to detect.After years of research and development,Electrical Capacitance Tomography(ECT)has emerged as an ideal means of non-destructive testing because of its non-invasive nature,high safety,stability and fast imaging speed.In this thesis,we investigate image reconstruction,flow pattern recognition and flow pattern parameter prediction in Electrical Capacitance Tomography technology with the 12-electrode ECT system as the object of study.The main contents are as follows.This article discuesses the composition and principles of Electrical Capacitance Tomography technology,and derives the positive and negative problem solutions of ECT technology through mathematical models and simulation experiments.First,the structure and working principle of the ECT system are introduced,and simulation experiments are built through the positive problem solving method.Then,the theoretical derivation of the inverse problem solution method is presented,and the advantages and disadvantages of the classical image reconstruction algorithm are analyzed qualitatively.Finally,the performance of four classical algorithms,namely,Linear Inverse Projection algorithm,Tikhonov regularization algorithm,Landweber iterative algorithm and Conjugate Gradient algorithm,is quantitatively analyzed through simulation experiments,with their common problems are pointed out.The above discussion provides the basis for the research of image reconstruction and flow pattern recognition methods in this thesis.When deep learning technology is used in image reconstruction,the image reconstruction accuracy is less satisfactory due to the influence of optimization methods.To address this problem,this thesis proposes an ECT image reconstruction algorithm that uses Sparrow Search Algorithm to optimize Deep Belief networks.The nonlinearity and ill-posedness of the measured capacitance values are analyzed,and an improved Deep Belief Network is used to process them.Considering that the Deep Belief Network has random weights and easily falls into local extreme value problems,the improved Sparrow Search Algorithm is introduced to optimize it,so as to establish the function relationship between the capacitance value vector and the image grayscale value.The experiments are compared with five image reconstruction algorithms,and the results show that the algorithm of this thesis can effectively improve the image reconstruction accuracy.It is difficult to find effective features in flow pattern identification,and noise has a greater impact on the recognition rate.To address this problem,this thesis proposes a fractal dimensional and multi-featured ECT flow pattern identification method.Firstly,the fractal characteristics of capacitance data are analyzed,the fractal dimension is calculated and the dimensional interval is divided,and the flow pattern is identified according to the flow pattern falling into different intervals.The experimental results show that the method can effectively distinguish the laminar flow,core flow and drop flow.On this basis,six effective features in the flow pattern data are analyzed and extracted,a 7-dimensional feature vector is formed with the fractal features.At the same time,considering the problem that the random parameters of the Support Vector Machine may reduce the classification efficiency,the Sparrow Search Algorithm is introduced to optimize it,and then the feature vectors are input into the Support Vector Machine to realize the flow pattern identification.The experiments are compared with six flow pattern identification methods,and the results show that the method can achieve accurate identification of six typical flow patterns with good robustness.The complexity of the prediction model has a large impact on the accuracy and time of the flow parameter prediction.To address this problem,this thesis proposes an ECT flow parameter prediction model using Principal Component Analysis technique and Particle Swarm Algorithm to optimize the Extreme Learning Machine.Firstly,the redundancy characteristics between the measured capacitance values are analyzed,and the Principal Component Analysis method is used to reduce the dimension of the sample data to obtain comprehensive factors that are independent from each other.Then,due to the convergence performance of the Extreme Learning Machine affected by the random parameters,the improved Particle Swarm Algorithm is used to optimize it,followed by inputting the reduced-dimensional composite factors into the optimized Extreme Learning Machine to realize the flow pattern identification.On this basis,known flow pattern parameters are introduced as a priori conditions for training the above model.Finally,the test samples are input into the prediction model to realize the flow pattern parameter prediction.The experiments are compared by four parameter prediction methods,and the results show that the method in this thesis can effectively improve the accuracy of flow pattern parameter prediction and the prediction time is shorter.