摘要
通过设计的三层BP神经网络对实验采集的11个声发射信号参数进行特征提取。首先,根据均方根误差确定隐含层中神经元的数量为12个,然后计算各声发射参数对表征裂纹信号灵敏度的大小,逐步删除灵敏度较小的声发射参数,以达到降低训练时输入信号维数的目的。最后确定选取幅度(X1)、振铃计数(X3)、能量(X6)、持续时间(X7)、时间消耗(X9)五个特征参数能够有效地识别金属拉深件裂纹。实验研究对于减少金属拉深件裂纹定位时的繁琐的参数计算具有重要意义。
It designed three layer BP neural network to extract characteristics parameters from the 11 acoustic emission signal. The first,according to the root mean square error,it determined the number of neurons in hidden layer is 12. And then,it calculates the size of the sensitivity of acoustic emission parameters on the characterization crack signal, and gradually removes various acoustic emission paranteter, reduces dimension of the input signal while training. Finally, it selects range (Xi), ringing count (X3), energy (X6), duration (XT), time consumption (X9), and the five characteristic parameters can effectively identify metal drawing crack. This experimental study has great significance of reducing tedious calculation parameters when locating the metal drawing parts crack.
出处
《机械设计与制造》
北大核心
2013年第9期126-128,共3页
Machinery Design & Manufacture
基金
国家科技型中小企业创新基金(09C26213201011)
关键词
拉深件
裂纹
声发射技术
BP神经网络
特征提取
Drawing Parts
Crack
Acoustic Emission
BP Neural Network
Feature Extraction