周俊a， 易良庭b， 王强b
（后勤工程学院 a.军事供油工程系；b.军事工程管理系,重庆 401331）
摘要: 针对现有声发射信号识别参数分析法的不足，提出利用小波变换特征提取与BP神经网络结合的声发射信号识别方法。利用db2小波对声发射信号进行6层分解，将细节特征空间的能量作为声发射信号特征向量；结合声发射信号特点设计和训练BP神经网络，通过实验确定隐层神经元数；用BP和RBF神经网络分别对腐蚀、裂纹和冷凝声发射信号进行分类测试。实验结果表明，提取的声发射信号特征具有较好的区分性能，BP网络在识别声发射信号方面优于RBF网络，这对储油罐安全状况的定量分析具有一定意义。
关键词: 声发射； 小波变换； 细节特征； BP网络； 识别
中图分类号： TP 391 文献标志码： A
Acoustic Emission Signal Recognition Based on Wavelet Transform and BP Neural Network
ZHOU Juna, YI Liangtingb, WANG Qiangb
(a.Department of Military Oil Supply Engineering； b.Department of Military Project Management，
Logistical Engineering University, Chongqing 401311, China)
Abstract: Against inadequateness of the acoustic emission signal analysis method by parameters,a new acoustic emission signal recognition method is proposed based on wavelet transform and BP neural network. Aoustic emission signal was decomposed to 6 layer by db2 wavelet,feature vector from acoustic emission signal is composed of the space energy based on 6 layer detail feature.BP neural network was designed and trained by taking the characteristics of acoustic emission signal into account, the BP network was trained by using the acoustic emission signal which pattern was known, hidden neurons numbers of BP network was determined by the experiment. Classify experiments by BP neural network and RBF neural network was done on corrosion emission signal, crack emission signal and condensation acoustic emission signal.The experimental results shows that the features extracted from acoustic emission signal has good distinguish performance,BP network is superior to RBF neural network in identification of the acoustic emission signal. There has a certain significance for quantitative analysis on oil storage tank safety situation.
Key words: acoustic emission; wavelet transform; feature; BP neural network; recognition
收稿日期： 20140311
基金项目： 军队后勤科研项目(油20080208);重庆市博士后科研项目(XM20120049).
作者简介： 周俊(1981—),男,高级工程师.
文章编号： 16726987（2015）03033808； DOI： 10.16351/j.16726987.2015.03.021
