导电结构缺陷电涡流检测稀疏特征提取与分类识别研究

发布时间：2018-04-29 21:21

本文选题：电涡流检测 + 缺陷　；参考：《昆明理工大学》2017年硕士论文

【摘要】：电涡流检测技术作为无损检测技术的一种,为导电结构材料提供了一种高效精确的缺陷检测方法,提高了导电结构材料在实际运用中的可靠性。实际运用过程中,良好的缺陷特征提取与分类识别方法是导电结构材料电涡流检测的关键,对缺陷信号进行特征提取的方法进行研究,以定量化地分类评估和研究缺陷的形状、大小以及材料等,一直都是电涡流检测以及反问题求解的难点。本文以导电结构材料作为电涡流检测对象,开展了针对导电结构材料电涡流检测的缺陷信号稀疏特征提取方法及缺陷定量化分类识别研究,具有重要的研究意义及应用价值。本文以电磁学理论及电涡流检测原理作为基础,分析推导了电涡流检测的关键技术及数学模型;借助ANSYS有限元仿真软件,针对导电结构材料的特性对电涡流检测有限元模型进行了搭建;针对不同缺陷尺寸及不同检测探头的设置方式进行了电涡流缺陷检测有限元分析计算,分析研究了计算结果同缺陷类型及检测探头形式变化之间的相互关联,总结各因素改变所带来的检测结果变化规律。针对导电结构材料电涡流检测信号,提出一种基于稀疏信号处理的拉格朗日乘子K奇异值分解(K-SVD)信号特征提取方法;以铝合金材料作为被测对象进行电涡流检测实验,设置不同的缺陷长度与缺陷深度,在改变激励频率的条件下获取检测得到的阻抗信号;对不同缺陷条件及不同激励频率的检测信号运用拉格朗日乘子K-SVD算法进行信号特征提取,计算上述条件改变下的特征分布结果,归纳不同缺陷与激励频率产生的探头阻抗变化规律,验证拉格朗日乘子K-SVD算法在缺陷信号特征提取方面的可行性及高效性。搭建SVM计算模型,针对铝合金材料电涡流检测缺陷特征,优化SVM算法模型;运用SVM分类识别方法对不同缺陷尺寸及不同激励频率下的缺陷特征进行定量化分类识别计算研究;归纳缺陷变化与频率变化产生的缺陷成功识别概率的变化规律;分析对比主成份分析法与拉格朗日乘子K-SVD算法在特征提取与分类识别的性能,验证拉格朗日乘子K-SVD算法的优越性。
[Abstract]:As a kind of nondestructive testing technology, eddy current testing provides an efficient and accurate method for defect detection of conductive structure materials, and improves the reliability of conductive structure materials in practical application. In the process of practical application, good defect feature extraction and classification recognition method is the key of eddy current detection of conductive structure materials. The method of feature extraction of defect signal is studied to quantitatively classify and evaluate and study the shape of defect. The size and material are always the difficulties of eddy current detection and inverse problem solving. In this paper, conducting structure materials are taken as the object of eddy current detection, and the methods of extracting the sparse feature of defect signals and the quantitative classification and recognition of defects for eddy current detection of conductive structure materials are carried out, which have important research significance and application value. Based on the theory of electromagnetism and the principle of eddy current testing, this paper analyses and deduces the key technology and mathematical model of eddy current testing, and makes use of ANSYS finite element simulation software. The finite element model of eddy current testing is built according to the characteristics of conductive structure materials, and the finite element analysis calculation of eddy current defect detection is carried out according to different defect sizes and different detection probe settings. The correlation between the calculation result and the type of defect and the change of the detection probe's form is analyzed and studied, and the rule of the change of the detection result caused by the change of each factor is summarized. Based on sparse signal processing, a signal feature extraction method based on Lagrange multiplier K singular value decomposition (SVD) is proposed for eddy current detection of conductive materials. By setting different defect length and depth, the impedance signal is obtained under the condition of changing the excitation frequency, and the detection signal with different defect condition and different excitation frequency is extracted by the Lagrange multiplier K-SVD algorithm. The results of characteristic distribution under the above conditions are calculated and the law of impedance variation of probe caused by different defects and excitation frequencies is summarized. The feasibility and high efficiency of Lagrange multiplier K-SVD algorithm in feature extraction of defect signals are verified. The SVM calculation model is built to optimize the SVM algorithm model for eddy current detection of aluminum alloy materials, and the quantitative classification and calculation of defect characteristics under different defect sizes and different excitation frequencies are carried out by using SVM classification method. Induces the change rule of the probability of defect recognition caused by the change of defect and frequency, analyzes and compares the performance of principal component analysis method and Lagrange multiplier K-SVD algorithm in feature extraction and classification recognition, and analyzes and compares the performance of principal component analysis method and Lagrange multiplier K-SVD algorithm in feature extraction and classification recognition. The superiority of Lagrange multiplier K-SVD algorithm is verified.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM24

【参考文献】