Being an inverse problem of electromagnetic fields, the quantitative inspection of pipeline cracks in remote field eddy current inspection ( RFECI) becomes an ill-posed problem for the lack of the prior constraints. The significant correlationship between the cracks and the features of the magnetic signals through the researches on the axisymmetric defects of the pipeline is demonstrated. A forward modeling, which can quantitatively map the pipeline defects to the features of the magnetic signals, based on back-propagation neural network (BPNN) is proposed. The high approximation accuracy and the good generalization ability of the forward modeling mean the effective prior knowledge and constraints for the quantitative inverse of the pipeline defects.%管道远场涡流裂纹缺陷检测的本质是电磁场反演问题,由于先验约束条件的不足,缺陷尺寸的定量检测成为一个无定解的不适定问题.提出了一种基于BP神经网络学习算法的管道远场涡流检测正演模型的设计方法,通过对轴对称缺陷管道模型的仿真研究,提取出与缺陷尺寸显著相关的关键磁场特征量,实现了从管道裂纹缺陷尺寸空间到磁场信号特征空间的非线性定量映射.经测试,正演模型对远场涡流特征信号具有良好的逼近精度和推广能力,可为管道轴对称裂纹缺陷的定量反演评估提供有效的先验知识和约束条件.
展开▼
