超声检测中金属裂纹多特征提取研究  

A study of multi-feature extraction for metal crack using ultrasonic testing

作　　者：樊萍 刘新宝[2] FAN Ping;LIU Xinbao(School of Information Science and Technology,Northwest University,Xi′an 710127,China;School of Chemical Engineering,Northwest University,Xi′an 710069,China)

机构地区：[1]西北大学信息科学与技术学院,陕西西安710127 [2]西北大学化工学院,陕西西安710069

出　　处：《西北大学学报（自然科学版）》2018年第4期521-526,共6页Journal of Northwest University（Natural Science Edition）

基　　金：国家自然科学基金资助项目(51371142);陕西省自然科学基金资助项目(2017JQ6079);陕西省教育厅自然基金资助项目(15JK693)

摘　　要：针对传统超声检测中大多采用单一特征致使裂纹检测不准确的问题,该文提出了基于多特征提取的金属裂纹检测方法。首先,在分析裂纹超声回波特点的基础上,利用小波包变换获取信号的局部时频信息;随后,通过定义相关系数、能量熵和模极大值等参数作为识别特征,结合k-近邻准则和决策级融合算法对金属裂纹进行识别分析。超声回波实测结果表明,利用该文提出的改进方法在显著提高金属裂纹识别率的同时,还能有效降低高斯白噪声影响。Usually, it is very difficult to identify the material crack exactly using the conventional ultrasonic testing （UT）. This is mainly attributed to the single feature adopted by UT. The multi-feature extraction method is proposed in the present study. Firstly, wavelet packet transform （WPT） of echoes is employed to obtain the local time-frequency resolution of them based on the analysis of crack echo properties. Then, the correlation coefficient, energy entropy and modulus maximum are defined as the features of crack identification. Consequently, the metal crack is identified with k-Nearest Neighbor （kNN） algorithm and decision fusion. Ultrasonic testing of metal crack showed that the proposed feature extraction can improve the classification performance significantly, while reducing the influence of Gausses white noise effectively.

关 键 词：超声检测 金属裂纹 多特征提取 小波包变换 

分 类 号：TN911.7[电子电信—通信与信息系统]