基于PSO-SVM的焊缝缺陷X射线检测  

Detection of defects in weld by X-ray based on PSO-SVM

作　　者：蔡晓龙[1] 穆向阳[1] 高炜欣[1] 魏巍 

机构地区：[1]西安石油大学陕西省钻机控制重点实验室,陕西西安710065 [2]西安西北石油管道公司,陕西西安710018

出　　处：《焊接技术》2013年第10期57-60,5,共4页Welding Technology

基　　金：陕西省自然科学基础研究计划项目(2010JQ8033)

摘　　要：为了提高X射线焊缝缺陷检测的实时性,本文研究了将粒子群优化(Particle Swarm Optimization,PSO)算法与支持向量机(Support Vector Machine,SVM)算法相结合应用于焊管焊缝缺陷检测的方法。该方法首先提取焊缝缺陷的特征描述,然后利用SVM算法进行焊缝缺陷的检测,过程中采用PSO算法优化SVM模型参数,最后将PSO-SVM和基于网格寻优的SVM分类方法进行了对比。试验结果表明,基于PSO-SVM的焊缝缺陷检测方法具有较高的识别精度,平均分类准确率达98%,且相对于后者其实时性提高了39.87%,这表明PSO-SVM方法能够有效地应用于焊管焊缝缺陷检测实时性的提高。To improve the real-time of X-ray weld seam defects detection, a method combined the Particle Swarm Optimization(PSO) algorithm and Support Vector Machine(SVM) algorithm was proposed in this paper. First, the features' description of weld seam defects were extracted from the X-ray images, then the SVM algorithm was adopted to detect weld defects, in this process the PSO algorithm was used to optimize the SVM parameters' selection. At last, the PSO-SVM method was compared with the SVM based on grid search method in their ability to detect defects. The experiment results showed that the PSO-SVM method had a higher accuracy with its average accuracy up to 98%. Compared with the SVM based on grid search method, it improved the real-time of X-ray weld defects detection to 39.87%. It indicated that the PSO-SVM could be used to improve the real-time of weld defects detection effectively.

关 键 词：粒子群优化 模型参数 缺陷检测 支持向量机 

分 类 号：TG441.7[金属学及工艺—焊接]