【作者】 李伟；

【导师】 陈国明； 顾心怿；

【作者基本信息】 中国石油大学 ， 机械设计及理论， 2007， 博士

【摘要】 本文受国家863青年基金项目“交流电磁场检测(ACFM)智能可视化技术研究”资助,针对ACFM技术缺陷智能识别和形状可视化的难题,通过理论分析和实验研究,系统地开展交流电磁场缺陷智能可视化检测系统开发的关键技术研究,在ACFM有限元仿真分析、缺陷定量识别以及形状可视化反演、ACFM缺陷智能可视化探头设计与优化、磁场信号软硬件处理技术、虚拟样机系统开发等方面均取得较大的研究进展,主要的研究成果总结如下:1 ACFM理论研究与数值模型建立结合ACFM检测原理和数值计算理论,引入矢量磁位A和标量势函数?,提出ACFM缺陷识别的理论模型。为了简化模型中缺陷附近电磁场分布的仿真分析和求解,引入有限元分析方法,针对分析对象和目的的不同,建立了3种不同的ACFM有限元分析模型:激励仿真模型、计算模型和探头运动仿真模型。激励模型可以准确的模拟检测中感应电磁场的分布规律,为探头设计和优化提供理论依据。计算模型能够完成稳定检测状态下缺陷上方磁通密度分布的分析和数值计算。探头运动仿真模型主要用于探头运动检测或大缺陷检测情况下磁场分布的分析和计算。以上3种专用的参数化仿真模型的建立,减少了ACFM电磁场仿真建模和计算的工作量,也为后续的研究奠定基础。2 ACFM缺陷智能识别与形状可视化算法研究鉴于目前ACFM缺陷识别方法中存在的人为误差大、无法实时识别和智能化程度低等方面的不足,对ACFM缺陷智能识别技术进行深入的研究。作为缺陷智能识别研究的基础,采用人工缺陷加工和仿真模拟相结合的方法,对缺陷进行分类制作,建立缺陷数据库,并在此基础上通过信号分析和特征提取,确定了用以标示缺陷信息,简化反演的缺陷信号特征量。分析当前ACFM缺陷判定方法的不足,借助相关检测,提出加入相位判据的直接判别法,实时判别缺陷的存在。统计分析缺陷数据库中部分缺陷的实验检测结果或仿真结果,基于缺陷尺寸对于磁场信号特征值影响的3条规律,提出基于插值的ACFM缺陷定量识别算法,为工程中含有大量检测数据的缺陷实时量化提供新途径。充分考虑缺陷尺寸和磁场特征分布之间存在的复杂非线性关系,引入非线性神经网络,建立基于BP神经网络和广义回归神经网络的缺陷定量识别模型,并对两种模型的量化精度以及推广性进行了对比分析。基于旋转感应磁场,提出了ACFM缺陷方位计算方法,为缺陷形状可视化奠定基础。针对目前国内外对于ACFM缺陷形状可视化技术研究的一些空白,通过大量的非规则形状缺陷的有限元仿真实验,分析由缺陷形状引起的磁场信号分布的变化。在此基础上,借鉴有限元分割思想,提出缺陷截面形状反演算法和缺陷表面形状反演算法以及缺陷三维重构方法,并进行仿真测试和误差分析,反演精度均超过90%。基于目标优化思想,融合缺陷形状反演算法和有限元仿真技术,设计缺陷可视化的智能实现流程,为缺陷形状可视化的计算机智能实现奠定基础。3 ACFM缺陷智能可视化检测系统研制针对ACFM磁场信号的特点,借助参数化激励仿真模型,探讨激励探头结构对感应电磁场的影响。基于旋转磁场理论,开发双U型ACFM正交激励阵列探头,为缺陷形状可视化和角度的检测提供必要的旋转匀强感应磁场,消除缺陷方向对检测灵敏度的影响。配合一维阵列检测线圈,设计一维ACFM智能可视化探头,为缺陷的智能可视化反演提供足够的磁场信息。考虑探头采集的磁场信号属于微弱、多干扰信号,对信号发生以及处理电路进行设计,实现信号的初步处理,为信号A/D转换和数字化信号处理奠定基础。结合A/D数据采集卡,在LabVIEW环境下设计信号采集与转换模块,本着简化硬件电路、降低仪器成本、提高系统参数化水平的目的,将部分硬件电路软件化,构建数字信号处理软件模块,设置专用滤波器,并参考正交锁相放大原理,开发互相关相位检测器,以正交激励信号为参考信号,实现信号的矢量测定,为后续的缺陷智能可视化反演奠定基础。鉴于MATLAB语言在数值分析和计算方面的独特优势,采用LabVIEW和MATLAB混合编程的方法编制缺陷智能可视化反演模块,开发ACFM缺陷智能可视化检测软件系统,实现缺陷的实时判别、自动量化及缺陷可视化描述。采用人工缺陷检测实验的测试方法,对开发的ACFM缺陷智能可视化检测系统的功能、精度以及不同环境下的检测性能进行测试,结果表明该系统量化精度不低于85%,反演所得缺陷外形可以较为准确的描述缺陷真实形状,而且对于水下结构物缺陷也能够实现智能可视化检测,量化精度也可达到85%。4 ACFM虚拟样机系统开发以交流电磁场检测原理为基础,利用数值仿真技术为主要设计手段,借助多领域集成仿真技术,对有限元仿真模型参数化设计与计算,仿真结果描述与特征提取,缺陷特征数据库维护,缺陷定量识别与可视化反演等ACFM相关的正、反问题进行模块设计和程序实现,开发ACFM缺陷智能可视化检测虚拟样机系统。该样机以ACFM仿真模型代替人工缺陷,以有限元计算结果作为原始磁场信号,自动完成数字信号的分析和特征提取、缺陷的智能定量识别以及形状反演和可视化描述,并提供与检测系统之间的数据通讯接口,利用检测系统调节虚拟样机中仿真模型的设计参数,利用虚拟样机的智能可视化子系统作为检测系统采集信号的后处理模块,有效提高ACFM缺陷检测系统的检测精度以及智能可视化水平。更多还原

【Abstract】 The dissertation is focused on the technology of defect intelligent recognition and visualization based on the alternating current field measurement (ACFM), including the FEM simulating analysis, the intelligent recognition and visualization of defect, the design and optimization for probe, electromagnetic signals processing and the development of virtual prototype system and so on, which is supported by‘863’Youth Fund of the High Technology Research and Development Program of China (No. 2002AA616060). The main works are summarized as follows:1 Study of the Theory and Numerical Model of ACFMBased on the principle of ACFM and the theory of numerical calculation, the magnetic vector A and the scalar potential function ? are introduced, and the theory model of ACFM defect recognition is presented. To analyze and calculate the electromagnetic field distribution around the defect in that theory model, the finite element analysis (FEA) method is inducted, and three FEA models are built for different subjects and aims. The induction model is used to simulate the induced electromagnetic field distribution in ACFM, which provides the theoretical foundation to design and optimize the probe. The calculation model is used to analyze and calculate the distribution of the magnetic flux density above the defect in the state of stable measurement. And in the circumstance of moving scan or the measured defect is bigger than the area of the induced field, the probe moving simulation model will be used to analyze and calculate the distribution of the magnetic field. These three specific models mentioned above make the workload of building model and analysis of ACFM electromagnetic field distribution decreasing effectively, and provide the foundation of the following study.2 Research on the Algorithm of Defect Intelligent Recognition and VisualizationConsidering the deficiency of involving man made error, difficulties of real-time recognition and low degree of automation and intelligence in the field of defects recognition, a deep research on the technique of defect intelligent recognition in ACFM is performed. As the basis of intelligent defect recognition, the defect database is built by combining the artificial defects producing and simulating. The characteristic vectors of defect signals are determined to reveal the defect information and simplify the algorithm by signal processing and feature extraction. Analyzing the shortcoming of defect discrimination methods in ACFM, a phase involved direct method used in defects real-time discrimination is presented on the basis of cross-correlation detection method. Based on the three rules describing the relationship between the magnetic field distribution and the size of defects, a interpolation algorithm of quantitative recognition of defects for ACFM is proposed, which presents a new method of defect real-time recognition in engineering. Considering the complex nonlinear relationship between the size of the defect and the distribution of the characteristic signals, the nonlinear neural network is introduced, and the BP neural network model and the generalized regression neural network (GRNN) model for defect recognizing in ACFM are built. Based on the rotational induced magnetic field, a method for calculating the direction of the defect in ACFM is proposed, which makes the foundation of defect shape visualization.The relationship between the distribution of the magnetic field signals and the defect shape is analyzed by lots of finite element numerical simulating experiments of arbitrary shape defects. On the basis of finite element segmentation method, the inversion algorithm of the section shape and surface shape of the defect and the reconstruction method of three dimensional profile of the defect are proposed. These algorithms and method are verified by simulation experiments, and the results show that the precision is higher than 85 percent. Combined the inversion algorithm of the defect shape with the finite element simulation technique, the intelligent process of defect shape visualization is designed by the optimization method.3. Development of ACFM Intelligent and Visual Recognition SystemConsidering the characteristic of the magnetic field signals in ACFM, the relationship between the structure of the inducer and the induced magnetic field is discussed by parameterized induced simulation model. On the basis of the principle of rotational magnetic field, a double U-shaped orthogonal inducer for ACFM is presented, which could clear up the unfavorable influence of the defect direction on the sensitivity of measurement. A one dimensional defect intelligent and visual recognition probe of ACFM is developed with one dimension array detecting coils, which provides enough magnetic information for intelligent defect recognition and visualization. Considering the weakness of the signals and disturbance of noises, by designing the signal producing and processing current circuits, the primary signal process is realized, which makes the foundation for A/D conversion and digital signal process.Using an A/D data acquisition card, by the LabVIEW language, the signal acquisition and conversion module is developed. And aiming at simplifying current circuits, and decreasing cost, parts of current circuits are realized by computer software, and the digital signal process software module is designed. On the basis of the principle of orthogonal lock-in amplification, a cross-correlation phase detector is developed, by which both the phase and amplitude of signal are measured. Due to the advantage of MATLAB software in the field of numerical calculation and analysis, the defect intelligent and visual recognition module is produced by the method of mixed language programming. Finally, the ACFM defect intelligent and visual recognition software system is built by combining these modules mentioned above, which can be used to realize the real-time defect discrimination, intelligent quantifying recognition, and visual description of defect shape.The ACFM defect intelligent and visual recognition system is tested by experiments of measuring artificial defects in the field of function precision and performance under some special circumstances. The results show that the quantifying precision of this system is about 90 percent, and the shape inversely calculated is similar to the real shape of the defect, and the defect under water can be also recognized by this system intelligently and visually, the precision about 85 percent.4 Design of Virtual Prototype System of ACFMBased on the principle of ACFM, researching on the forward and inverse problems such as design and calculation of the FEA parametrical model, simulation results process and feature extraction, maintenance of the defect characteristic database, and defect quantifying recognition and visual inversion and so on, a virtual prototype system for defect intelligent recognition and visualization based on ACFM is built by the numerical simulation technique by C++ BUILDER language. Aiming to adjust the parameters of virtual prototype and test the function of measurement system, an interface subsystem is designed to link the measurement system with the virtual prototype. In addition, the virtual prototype could be used as a post-processing module for signals acquired by the probe of measurement system.更多还原