基于EM聚类的敲击检测缺陷识别研究  

Research on Knock Detection Defect Recognition Based on EM Clustering

作　　者：柳志高 李艳军[1] 曹愈远[1] 赵连州 LIU Zhi-gao;LI Yan-jun;CAO Yu-yuan;ZHAO Lian-zhou(Civil Aviation College,Nanjing University of Aeronautics and Astronautics,Nanjing 211000,China)

机构地区：[1]南京航空航天大学,江苏南京211000

出　　处：《航空计算技术》2022年第2期53-57,共5页Aeronautical Computing Technique

基　　金：航空科学基金项目资助(2020033052001)。

摘　　要：复合材料的优点是重量轻和装配简单,近年来复合材料在航空工业中的使用量正在不断地增加,但航空安全一直是倍受关注的问题,因此对航空复合材料的无损检测就显得尤为重要。以研究脱粘缺陷为中心,将敲击检测的数据作为样本并使用EM聚类算法对其进行分类,并估计高斯混合模型的参数,建立正常数据和缺陷数据的两个高斯分布模型,用于敲击检测过程中介于缺陷与正常之间的临界点数据的识别研究。实验仿真结果表明,基于高斯分布的EM聚类算法分类的准确率达到了99.958%,分类的重叠度为0.384,可见模型可以用于实际工程应用中敲击检测得到的介于缺陷与正常之间的临界点数据的判别,对提高敲击检测故障诊断智能化水平具有较高的参考价值。In recent years, the advantages of composites are light weight and simple assembly, so the use of composites in the aviation industry is increasing, but aviation safety has always been our concern, so the nondestructive testing of aviation composites is particularly important.This paper focuses on the research of debonding defects, takes the knock detection data as samples, classifies them using EM clustering algorithm, estimates the parameters of Gaussian mixture model, and establishes two Gaussian distribution models of normal data and defect data, which are used to identify the critical point data between defect and normal in the process of knock detection.The experimental simulation results show that the classification accuracy of EM clustering algorithm based on Gaussian distribution is 99.958% and the classification overlap is 0.384.It can be seen that the model can be used to distinguish the critical point data between defect and normal obtained from knock detection in practical engineering application, and has high reference value for improving the intelligent level of fault diagnosis of knock detection.

关 键 词：复合材料 缺陷识别 敲击检测 EM聚类 高斯分布 

分 类 号：V261[航空宇航科学与技术—航空宇航制造工程]