应用激光, 2018, 38 (4): 656, 网络出版: 2018-10-06
焊接缺陷动态磁光成像检测与分类研究
Detection and Classification of Dynamic Magneto-optical Images for Welding Defects
激光焊接 缺陷检测 磁光成像 特征提取 神经网络 laser welding defect detection magneto-optical imaging feature extraction neural network
摘要
针对激光焊接过程中产生的焊缝表面缺陷难以检测与识别的问题, 研究一种交变磁场激励下采集焊缝表面缺陷的磁光成像动态无损检测方法。分析基于法拉第磁光效应的焊接缺陷磁光成像机理, 使用磁光传感器动态采集了高强钢焊接缺陷磁光图像。使用灰度共生矩阵法分析并提取了图像的纹理特征, 使用LM(Levenberg-Marquardt)算法改进型BP(Back Propagation)神经网络建立了焊接缺陷分类模型。实验结果表明, 所提方法可以识别高强钢焊件中的焊缝缺陷(裂纹、未熔透、凹陷), 缺陷分类模型的整体识别率可达到96.75%。有效实现了焊缝表面缺陷的检测与分类。
Abstract
A dynamic magneto-optical imaging nondestructive detection method of weld surface defects under alternating magnetic field excitation was studied to solve the problem of detection and recognition of the surface defects in the laser welding. The mechanism of magneto-optical imaging based on Faraday magneto-optical effect was analyzed. The weld defects of high strength steel were collected by using a magneto-optic sensor. The gray scale co-occurrence matrix method was used to analyze and extract the texture features of images. The BP(back propagation)neural network improved by the LM(Levenberg-Marquardt)algorithm was used to establish the weld defect classification model. The experimental results show that the proposed method can identify the weld defects(cracks, partial penetration and sags)in high-strength steel weld pieces, and the overall recognition rate of defect classification model can reach 96.75%. The detection and classification of weld surface defects can be realized effectively.
郑俏俏, 高向东, 代欣欣. 焊接缺陷动态磁光成像检测与分类研究[J]. 应用激光, 2018, 38(4): 656. Zheng Qiaoqiao, Gao Xiangdong, Dai Xinxin. Detection and Classification of Dynamic Magneto-optical Images for Welding Defects[J]. APPLIED LASER, 2018, 38(4): 656.