基于BP神经网络的漏磁定量化检测

Quantitative testing of magnetic flux leakage based on BP neural networks

漏磁检测是由铁磁材料制作的兵器部件的常用无损检测方法之一,检测中的难点是根据被测漏磁信号反演缺陷的几何参数。将BP神经网络应用于漏磁信号的反演中,对神经网络进行训练,建立了漏磁信号与缺陷几何参数之间的数学模型,利用测量漏磁信号和仿真数据对模型进行了检验。试验结果表明,BP神经网络能根据漏磁信号精确地预测缺陷的几何参数,为漏磁定量化检测提供了一种可行的方法。

The magnetic flux leakage （MFL） method is generally employed in the nondestructive evaluation （NDE） of weapon for ferromagnetic materials, the difficulty is to invert the defect geometry parameters according to the measured MFL signals. BP neural networks are applied in the inversion of MFL signals. The mathematic models between the MFL signals and the defect geometry parameters are built based on neural networks. The models are then tested using both the measured MFL signals and the simulated data sets. The experimental results indicate that the BP neural networks can accurately predict the defect geometry parameters according to the MFL signals, the viable method is offered for MFL quantitative testing.