基于支持向量机和特征选择的超声缺陷识别方法研究  

Support Vector Machine and Chaos-genetic Optimization Based Ultrasonic Flaw Identification

作　　者：刘清坤[1] 阙沛文[1] 郭华伟[1] 宋寿鹏[1] 

机构地区：[1]上海交通大学,上海200030

出　　处：《中国机械工程》2006年第1期9-12,共4页China Mechanical Engineering

基　　金：国家863高技术研究发展计划资助项目(2001AA602021)

摘　　要：缺陷的自动识别在管道缺陷超声无损检测和评估中具有重要意义。提出了一个新的超声缺陷自动识别系统。该系统采用小波包分解提取超声信号的特征,采用混沌遗传算法来消除冗余和不相关的特征,并采用支持向量机(SVM)来对缺陷进行识别。为了验证系统的有效性,在实验室内作了大量的超声检测实验。实验结果表明,系统对管道缺陷具有较高的分类性能。Automatic identification of flaws is very important for ultrasonic nondestructive testing and evaluation of pipelines. A novel automatic identification system of flaws was presented. Wavelet packet decomposition （WPD） was applied to feature extraction of ultrasonic signals, chaos-genetic optimization to eliminate redundant and irrelevant features and support vector machine to perform the identification task. To validate this system, some experiments are performed and the results show the proposed system has very high identification performance for pipeline flaws and outperforms other discussed methods.

关 键 词：超声检测 缺陷识别 混沌遗传算法 特征选择 支持向量机(SVM) 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]