Application of principal component analysis in defect identification using pulsed eddy current method
-
摘要: 在钢结构脉冲涡流缺陷识别中，通常采用信号的峰值幅度、过零时间、主峰面积等特征参数对缺陷进行表征。但上述参数相互关联，存在一定的信息冗余，增加了数据分析量及信息筛选难度，进而影响了缺陷识别的效率。针对上述问题，采用主成分分析法对脉冲涡流信号的6个特征参数进行降维处理，构造了一个主成分特征，减少了信息冗余；将上述主成分特征输入Logistic分类器，实现了对钢结构减薄缺陷的准确识别。结果表明：主成分分析法可以在确保缺陷识别准确率的情况下，有效减少分类器处理的数据量，提高缺陷识别效率。
-
关键词:
- 主成分分析 /
- 脉冲涡流 /
- 缺陷识别 /
- Logistic分类
Abstract: In the pulsed eddy current testing (PEC) of steel structures, multiple characteristics such as peak amplitude, zero-cross time, and peak area etc., are usually used for defect characterization. However, information redundancy existing among the above characteristics increases the amount of data for the analysis and the difficulty of information filtering; this could influence the efficiency of defect identification. In order to reduce information redundancy, principal component analysis (PCA) was employed to compress the six characteristics into one principal component characteristic which was then used as input for Logistic classifier to identify thinning defects in steel structure. The results show that PCA can effectivly reduce the amount of data processed by classifier and improve the efficiency of defect identification while ensuring accuracy. -
-
[1] 沈功田, 李建, 武新军. 承压设备脉冲涡流检测技术研究及应用[J]. 机械工程学报, 2017, 53(4): 49-58. [2] 武新军, 张卿, 沈功田. 脉冲涡流无损检测技术综述[J]. 仪器仪表学报, 2016, 37(8):1698-1712. [3] BOWLER J R, HARRISON D J. Measurement and calculation of transient eddy-currents in layered structures [J].Review of Progress in Quantitative Nondestructive, 1992,28(1):241-248.
[4] YE C, SU Z, ROSELL A, et al. Decay time method with PEC and MR sensor for linearly measurement of material electrical conductivity[J]. NDT & E International, 2019, 102:169-174.
[5] SOPHIAN A, TIAN G Y, TAYLOR D, et al. A feature extraction technique based on principal component analysis for pulsed Eddy current NDT [J]. NDT&E International, 2003, 36(1):37-41.
[6] 克劳斯·巴克豪斯. 多元统计分析方法[M]. 上海:上海人民出版社, 2009. [7] 张文彤. SPSS统计分析基础教程[M]. 北京:高等教育出版社, 2017.
计量
- 文章访问数: 0
- HTML全文浏览量: 0
- PDF下载量: 0
下载:
