基于Faster R-卷积神经网络的金属点阵结构缺陷识别方法  

Internal Defect Detection of Metal Three-dimensional Multi-layer Lattice Structure Based on Faster R-CNN

作　　者：张玉燕[1,2] 李永保 温银堂[1,2] 张芝威 ZHANG Yuyan;LI Yongbao;WEN Yintang;ZHANG Zhiwei(School of Electrical Engineering,Yanshan University,Qinhuangdao 066004,Hebei,China;Hebei Province Key Laboratory of Measuring and Testing technologies and Instruments,Yanshan University,Qinhuangdao 066004,Hebei,China)

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004 [2]燕山大学测试计量技术及仪器河北省重点实验室,河北秦皇岛066004

出　　处：《兵工学报》2019年第11期2329-2335,共7页Acta Armamentarii

基　　金：河北省自然科学基金项目(E2017203240)

摘　　要：采用增材制造技术制备的金属三维点阵结构可能存在裂纹、未熔合、断层等缺陷,导致金属点阵结构的结构-功能性能下降,为此提出一种金属三维多层点阵结构内部缺陷的检测方法。在Faster R-卷积神经网络架构基础上设计特征提取网络,结合工业CT扫描图片,对得到的断层灰度图像中缺陷部位进行快速、准确、智能检测识别和定位。实验验证结果表明,对金属三维多层点阵结构样件的内部典型缺陷识别率达到99. 5%.The cracks,incomplete fusion,faults and other defects may exist in the metal three-dimensional lattice structure prepared by additive manufacturing technology,which lead to the decline of structure-functional performance of metal lattice structure. A Faster R-CNN-based internal defect detection method is proposed for metal three-dimensional multi-layer lattice structure. A feature extraction network is designed on the basis of the Faster R-CNN network architecture. It makes the defects in the obtained gray-scale image and the CT scanning image be detected and positioned quickly,accurately and intelligently. The experimental results show that the recognition rate of the typical internal defects of metal three-dimensional multi-layer lattice structure sample is 99. 5%.

关 键 词：金属点阵结构 缺陷识别 无损检测 CT扫描图像 Faster R-卷积神经网络 

分 类 号：TG115.281[金属学及工艺—物理冶金]