基于改进FCM和PSO-SVM的焊接缺陷识别  

Welding Defect Recognition Based on Improved FCM and PSO-SVM

作　　者：穆晨光 王海登 符浩[1] 边传新 史新鑫 MU Chen-guang;WANG Hai-deng;FU Hao;BIAN Chuan-xin;SHI Xin-xin(Luoyang Ship Material Research Institute,Henan Luoyang 471000,China)

机构地区：[1]洛阳船舶材料研究所,河南洛阳471000

出　　处：《失效分析与预防》2024年第3期179-185,共7页Failure Analysis and Prevention

摘　　要：为实现海洋工程钢结构件焊接接头缺陷的客观、智能化分类,本文以其数字射线检测图像作为研究对象,进行基于改进的模糊C均值聚类算法(FCM)和粒子群优化支持向量机(PSO-SVM)的缺陷识别研究。首先,基于限制对比度直方图均衡化去除原始图像中干扰噪声,引入像素点加权系数ω改进FCM进行图像分割;然后,基于灰度共生矩阵提取图像纹理特征,利用主成分分析法进行特征数据降维,将粒子群优化与支持向量机分类相结合进行参数寻优,建立纹理特征与缺陷类型间的连续变量分类模型;最后,以多人工综合完全正确的评价结果验证缺陷识别模型的有效性和准确性。结果表明:所训练的识别模型准确率为96.11%,经验证其识别准确率约为95.2%。与未经限制对比度自适应直方图均衡化(CLAHE)增强的模型、反向传播(BP)神经网络模型对比,该模型可以很好地实现常见缺陷的识别,且误差小,可应用于船用钢数字射线焊接缺陷识别领域。In order to realize the objective and intelligent classification of the defects of welded joints of marine engineering steel parts,digital radiographic images were taken as the research object to study the defect identification based on improved Fuzzy C-means(FCM)clustering-algorithm as well as particle swarm optimization-support vector machine(PSO-SVM).Firstly,the interference noise in original images was removed based on contrast limited adaptive histogram equalization(CLAHE),and the pixel weighting coefficientωwas introduced to improve FCM clustering-algorithm for image segmentation.Then,the texture features of images were extracted based on gray level co-occurrence matrix,the dimensionality of feature data was reduced by principal component analysis,and the parameters were optimized by combining PSO and SVM classification.As a result,a continuous variable classification model between texture features and defect types was established.Finally,the effectiveness and accuracy of the defect identification model were verified by manual statistics of the completely correct evaluation results.The results show that the accuracy of the established recognition model is 96.1%,and after verification,the recognition accuracy of this model is 95.2%.Compared with the model without CLAHE enhancement and the back propagation(BP)neural network model,this model can well realize the identification of common defects with small error,which can be applied to the digital radiographic identification field of welding defects of marine steel.

关 键 词：改进FCM 纹理特征 粒子群算法 支持向量机 缺陷识别 

分 类 号：TG115.28[金属学及工艺—物理冶金]