基于GA-BP神经网络的ACFM实时高精度裂纹反演算法

摘要

It is hard to achieve a real-time and high-precision cracks inversion for alternating current field measurement (ACFM) based on traditional characteristic signals. In this paper, based on the finite element method (FEM) model of e-lectromagnetic coupling ACFM probe, the energy spectrum and phase threshold determination methods were presented to ob-tain the crack characteristic signals in real time. The real-time and high-precision cracks inversion system for ACFM was set up and verified by artificial cracks experiment. The length and depth of cracks were calculated using the characteristic signals obtained from experiments based on the genetic algorithm and back propagation neural network( GA-BP) real-time and high-precision cracks inversion algorithm. The results show that the FEM model of electromagnetic coupling ACFM probe can sim-ulate the characteristic signals perfectively, the energy spectrum and phase threshold determination method can obtain the crack characteristic signals in real time, the GA-BP neural network can realize the inversion of the length and depth of crack perfectly and the relative error of inversion accuracy is less than 10%.%针对传统交流电磁场检测( ACFM)特征信号难以实现缺陷高精度实时反演的问题，在电磁耦合ACFM探头有限元模型分析的基础上，引入能量谱和相位阈值判定方法实时获取裂纹特征信号，建立裂纹实时反演实验系统并进行裂纹检测实验，基于加入遗传算法的BP神经网络( GA-BP)建立的ACFM实时高精度裂纹反演算法对实验得到的裂纹特征信号进行长度和深度的反演。结果表明：电磁耦合ACFM探头有限元模型可较好地仿真裂纹特征信号；采用能量谱和相位阈值判定方法能够实时获取裂纹特征信号；GA-BP神经网络能够实现裂纹长度和深度的反演，反演精度误差不超过10%。