基于局部离群因子与隔离森林的激光超声缺陷检测  

Laser Ultrasonic Defect Detection Based on Local Outlier Factor and Isolated Forest

作　　者：李阳[1] 朱文博[1] 静丰羽 叶中飞 马云瑞 周洋[1] 邹云[1] LI Yang;ZHU Wenbo;JING Fengyu;YE Zhongfei;MA Yunrui;ZHOU Yang;ZOU Yun(School of Mechanical and Power Engineering,Zhengzhou University,Zhengzhou 450001,China;POWERCHINA Henan Electric Power Equipment Co.,Ltd.,Luohe 462000,China;State Grid Henan Electric Power Research Institute,Zhengzhou 450052,China)

机构地区：[1]郑州大学机械与动力工程学院,河南郑州450001 [2]中国电建集团河南电力器材有限公司,河南漯河462000 [3]国网河南省电力公司电力科学研究院,河南郑州450052

出　　处：《郑州大学学报（工学版）》2025年第1期105-112,共8页Journal of Zhengzhou University（Engineering Science）

基　　金：河南省重点研发与推广专项(222102220025)。

摘　　要：针对激光超声(LU)缺陷检测中最大振幅图存在伪像的问题,结合主成分分析(PCA)和两种无监督的机器学习算法局部离群因子(LOF)与隔离森林(IF),以实现对LU数据的无监督异常检测。首先,利用PCA算法对LU数据进行降维处理,减轻了LU数据的复杂度;其次,利用LOF算法和IF算法进行了数据异常值的识别分析,并利用累积分布函数和核密度估计确定异常值的阈值大小;最后,对比了LOF算法、IF算法以及最大振幅图的检测结果。结果表明:LOF算法有更优的缺陷识别精度和更低的误判率。In response to the issue of artifacts in the maximum amplitude images in laser ultrasonic(LU)defect detection,principal component analysis(PCA)was integrated with two unsupervised machine learning algorithms including local outlier factor(LOF)and isolation isolated forest(IF)to perform unsupervised anomaly detection on LU data.Firstly,the PCA algorithm was used to reduce the dimensionality of the LU data,simplifying its complexity.Secondly,the LOF and IF algorithms were employed to identify outliers in the data,and the thresholds for these outliers were determined using the cumulative distribution function and kernel density estimation.Finally,a comparison of the detection results from the LOF,IF algorithms,and the maximum amplitude images revealed that the LOF algorithm offered superior defect detection precision and a lower false positive rate.

关 键 词：激光超声 缺陷检测 主成分分析 局部离群因子 隔离森林 铝合金 

分 类 号：TH16[机械工程—机械制造及自动化] TB551[理学—物理]