摘要
介绍了漏磁无损检测的原理和漏磁场数学模型,提出了用小波神经网络对裂纹形状重构。在仿真试验中,用训练样本对小波神经网络进行训练逼近裂纹形状,训练采用随机梯度下降算法。训练样本来自漏磁场数学模型数据和测量的火炮漏磁信号,用测量数据重构裂纹形状。小波神经网络是多分辨率逼近,通过改变网络的分辨率控制输出精度。结果表明。
The principle and the mathematical model for magnetic flux leakage (MFL) nondestructive inspection were introduced, the use of wavelet neural networks (WNN) for the reconstruction of flaw profiles was presented in this paper. In the simulation experiment, WNN is first trained to approximate the flaw profiles using the trained samples from both the mathematical model data and the measured gun's MFL signals. A stochastic gradient descent algorithm was adopted in the training procedure. The trained WNN was then employed to predict the flaw shape from the measured MFL signals, while WNN is a multi-resolution approximation. The accuracy of estimated flaw profiles is controlled by adapting the number of resolution. The result indicates that the WNN approach can accurately reconstruct flaw profiles.
出处
《兵工学报》
EI
CAS
CSCD
北大核心
2005年第3期379-382,共4页
Acta Armamentarii
基金
国家自然科学基金资助项目 (5 0 175 10 9)
关键词
材料检测与分析技术
漏磁
裂纹重构
小波神经网络
多分辨率逼近
磁偶极子
火炮身管
measuring and analyzing for material
magnetic flux leakage
flaw profiles reconstruction
wavelet neural network
multi-resolution approximation
magnetic dipole
gun barrel
