【作者】 何世钧；

【导师】 王化祥；

【作者基本信息】 天津大学 ， 检测技术与自动化装置， 2005， 博士

【摘要】 电容层析成像(ECT)技术是基于电容敏感机理的过程层析成像(PT)技术。ECT系统具有非侵入、响应速度快、成本低等优点,可提供封闭管道及容器内部二维/三维可视化信息,在工业过程参数可视化检测应用中是一种非常有发展潜力的技术。本文的研究内容主要包括传感器的结构参数优化设计,图像重建算法研究以及硬件系统设计。主要研究内容和结果如下:1、编写了基于Matlab的ECT系统仿真软件包。软件包可自动建立8电极、12电极和16电极的二维/三维传感器模型;可计算灵敏度系数分布,绘制二维/三维灵敏度系数分布图。逆问题部分实现了反投影算法、动态滤波反投影算法、RBF神经网络算法以及SVM算法。2、传感器的结构参数优化设计。对传感器敏感场的灵敏度分布进行了研究,分析了电极数目、结构参数对敏感场的影响,并以此对传感器的结构参数进行了优化设计;以四面体作剖分单元构成三棱柱六面体成像单元,实现了用线性插值函数的三维有限元剖分,建立了ECT传感器的三维有限元模型;给出了传感器三维敏感场的灵敏度分布,分析了其轴向响应特性,得出了介质分布沿轴向变化时的三维仿真计算结果。3、动态滤波反投影算法的提出。基于敏感场分布的不均匀性,对滤波反投影算法进行了修正,提出了动态滤波反投影算法。4、基于RBF网络的图像重建。对RBF网络在图像重建中的应用进行了探讨。5、基于支持向量机的图像重建研究。提出了基于C-支持向量分类机的图像重建算法;提出了一种基于支持向量机的四层神经网络算法用于三相流图像重建;实现了基于支持向量机的三维图像重建,进行了仿真实验。6、硬件系统设计。构建了一套ECT系统样机,设计方案中应用了嵌入式技术和CAN总线等技术,获得了大量的实验数据,为进一步的研究奠定了基础。更多还原

【Abstract】 Electrical capacitance tomography(ECT)based on capacitance sensing array isa kind of process tomography (PT). It has the advantages of being non-intrusive,fast in response and low in cost. It can acquire visible 2D/3D distributioninformation of closed pipeline or vessel. The visible technique is of a greatdeveloping potential in industrial process parameter measurement .Although ECT technique has achieved great progress in the recent twenty years,and various performance indexes have been improved. But this falls short of actualapplications in industrial processes, many foundational researches need also to bedone in the future.In this thesis, the author studied mainly three aspects, i.e. optimum design ofstructure parameter of capacitance sensing array electrode, reconstruction algorithmand hardware design of ECT system. The following achievements have beenobtained:1 The development of the simulation software package of ECT systemThe simulation software package of ECT system based on Matlab has beendeveloped, which can be used in 2D/3D simulation according to demands. Itincludes forward problem and inverse problem. The 2D/3D models of 8-electrode,12-electrode and 16-electrod can be constructed automatically in solving forwardproblem. To counter different model, the dissection and finite element computationcan be done. The 2D/3D sensitivity distribution (map) of ECT sensors can beobtained. The inverse problem is called as reconstruction algorithms, which includeback projection method and active filter back projection method improved by author,and moreover, the simulation of images reconstruction is performed using RBFneural network and SVM method2 The optimum design of structure parameter of sensing array electrodeThe sensing filed in ECT system is called as ‘ soft-filed ’ , which isinhomogeneous,‘ill posed’ ,nonlinear and mainly relies on the structure parameterof array electrode. The sensitivity distribution is calculated by using the 2D finiteelement method (FEM). The effect of the number and the structure parameters ofarray electrode on the field uniformity are analyzed . A set of optimum parametersof sensing array electrode based on this method is obtained .In order to study ECTsensor in detail, a 3D capacitance model based on the FEM has been developed.Four-node tetrahedron elements are used for 3D meshing, six-node pentahedronelements which are made up of four-node tetrahedron elements are used as imageconstruction elements, in this way an linear interpolation function can be used inthis method. The effect of medium distribution in the axial direction is analyzed byusing calculating results of sensitivity distribution, and 3D simulation results aregiven.3 The presentation of the active filter back projection methodBased on inhomogenous of sensitivity distribution distribution the active filterback projection method is presented by improving filter back projection method.Simulation results show that the new method is superior to the classical filter backprojection method4 The applying the RBF neural network to the image reconstructionThe ECT is a typical nonlinear mapping problem, and the mapping model isdifficult to be described analytically. A neural network offers a general frameworkfor representing nonlinear function mapping from several input variables to severaloutput variables. In fact the image reconstruction based on neural network isprovided with nonlinear mapping from sampling capacitance data to pixel values.The RBF neural network provided a good properties of approximation, classifyand convergence. There is a liner relationship between network weight value andoutput value. It has no local minimum problem, and it is an optimization networkcompared with other forward networks. The applying the RBF neural network to thereconstruction is studied5 The investigating image reconstruction based on Support VectorMachine(SVM)The capacitance tomography is a typical small samples and nonlinear mappingproblem. Support vector machines (SVM) is based on the special small samplestheory with strong generalization ability, and is selected as an optimal theory forsmall samples classify problem.The ECT image reconstruction algorithms based on C-SVM is proposed. In thispaper a novel training method is proposed to improve the efficiency of C-SVMclassifier by selecting active penalty parameters;an image reconstruction algorithmbased on C-SVM is proposed by using four-layers neural network for three-phaseflow. The 3D image reconstruction algorithm is implemented by simulation.6 Hardware design of ECT systemAn ECT system is designed and developed successfully. The embeddedtechnique and CAN bus are applied in the design. The system possesses the stability,flexibility,expansibility and can reconstruct relatively good images.更多还原