基于改进BP神经网络的微裂纹漏磁定量识别  

Quantitative Identification of Microcracks Through Magnetic Flux Leakage Based on Improved BP Neural Network

作　　者：邱忠超[1,2] 张卫民[1] 张瑞蕾 马春红 

机构地区：[1]北京理工大学机械与车辆学院,北京100081 [2]机械科学研究总院先进制造技术研究中心,北京100083 [3]河北环境工程有限公司,河北承德067000

出　　处：《东北大学学报（自然科学版）》2016年第12期1759-1763,共5页Journal of Northeastern University(Natural Science)

基　　金：国家自然科学基金资助项目(51275048)

摘　　要：漏磁检测是铁磁材料常用的无损检测方法之一,定量识别是指通过检测到的漏磁信号识别裂纹的尺寸.采用主成分分析和优化神经网络相结合的建模方法,建立了微裂纹宽度与深度的预测模型.主成分分析去除了数据相关性,减小了输入样本维数,显著简化了网络结构;遗传算法优化的BP神经网络(GA-BP神经网络)可以有效地防止搜索过程中陷入局部最优解.通过基于磁偶极子模型的理论计算与人工刻槽微裂纹漏磁检测实验两种途径验证了该算法在微裂纹定量识别中的应用,为裂纹发展阶段的早期定量识别技术奠定了一定的基础.Magnetic flux leakage detection is one of NDT methods for ferromagnetic materials.Quantitative identification is to identify the crack size through obtaining magnetic flux leakage signals. By combining principal component analysis（ PCA） and neural network,a model was established to predict width and depth of the micro crack. The principal component analysis removed the data correlation and reduced the dimension of the input samples,so it can significantly simplify the network structure. BP neural network optimized by genetic algorithm（ GA-BP neural network） can prevent the search process from running into the local optimal solution. Based on the theoretical calculation of magnetic dipole model and experiment on the artificial cracks, the algorithm applied in the quantitative recognition of microcracks was verified,which may lay the foundation for the early quantitative recognition technique of crack development stage.

关 键 词：漏磁检测 主成分分析 GA-BP神经网络 微裂纹 定量识别 

分 类 号：TG115.28[金属学及工艺—物理冶金]