摘要
电磁声发射技术是一种新型的无损检测技术,通过对金属部件进行电磁加载会在裂纹处激发出声发射信号,并利用这一现象实现对金属材料的无损检测。本文分析了电磁声发射技术的基本原理与实现过程,采用一种基于波形分析的神经网络模式识别方法,利用小波包变换提取出电磁声发射信号波形的识别特征参数,建立了由10个输入单元、18个隐含单元和单输出组成的人工神经网络识别系统。为了克服BP神经网络收敛速度慢的缺点,提出了一种输入单元数目可变的神经网络改进方法,实验表明该系统能够对有无裂纹板进行快速、准确的识别。
Electromagnetically induced acoustic emission（EMAE） technique is a new nondestructive testing（NDT）.It does nondestructive detection with the effect of dynamic electromagnetic loading to generate a stress field stimulating stress waves from the defects.The principle and implementation procedure of the EMAE is analyzed.It adopts the neural network recognition method based on wave analysis.The characteristic parameters of EMAE signal are extracted using wavelet packet transform.The recognition system of back-propagation（BP） network consists of 10 input elements,18 hidden elements and single output.In order to overcome the shortcoming of low constringency speed,this paper proposes a kind of neural network recognition with adaptive number of neurons on the input layer method.The experiment results show it can identify the crack in the metal plate quickly and accurately.
出处
《电工技术学报》
EI
CSCD
北大核心
2012年第4期18-23,共6页
Transactions of China Electrotechnical Society
基金
supported by the National Natural Science Foundation of China(51077036)
the Natural Science Foundation of Hebei Province(E2012202048,E2011202040)
the Research and Development Project of Seience and Technology of Hebei Province(11215648)
关键词
电磁声发射
信号处理
神经网络
信号识别
Electromagnetically induced acoustic emission
signal processing
neural network
signal recognition
