焊接缺陷的磁光成像小波多尺度识别及分类

发布时间：2018-04-11 18:42

本文选题：磁光成像 + 焊接缺陷　；参考：《光学精密工程》2016年04期

【摘要】：针对焊缝微小凹陷、未熔合和焊偏等焊接缺陷,提出了基于磁光成像无损探伤的小波多尺度边缘提取算法及主成分分析-误差反向传播神经网络(PCA-BP)缺陷分类模型;研究了焊件表面及近表面缺陷的可视化无损检测及分类方法。首先,通过对焊件施加感应磁场,利用法拉第磁致旋光原理构成磁光传感器,获取焊接缺陷磁光图像。然后,针对焊接缺陷磁光图像存在噪声干扰、对比度低且成像背景复杂等特征,基于小波模极大值的多尺度边缘信息融合方法,设计了具有高抗噪性的缺陷边缘检测算法。最后,通过PCA法对磁光图像列方向灰度变量进行预处理,得到能表征95%磁光图像列方向灰度变量信息的256个特征点作为输入特征量,构建了三层BP神经网络模型,对焊接缺陷样本进行分类。试验结果表明,所提方法能准确识别微小凹陷、未熔合和焊偏等焊接缺陷,模型分类准确率可达90.80%。
[Abstract]:Aiming at welding defects such as micro-depression, non-fusion and welding deviation, a wavelet multi-scale edge detection algorithm based on magneto-optic imaging nondestructive testing and a PCA-BP-based defect classification model based on principal component analysis (PCA) and error back-propagation neural network (PCA-BP) are proposed.Visual nondestructive testing and classification method for surface and near surface defects of welded parts are studied.Firstly, the magneto-optic sensor is constructed by applying magnetic field and Faraday magneto-optic principle to obtain the magneto-optic image of welding defect.Then, aiming at the characteristics of noise interference, low contrast and complex imaging background in the magneto-optical image of welding defects, a high noise resistant defect edge detection algorithm is designed based on wavelet modulus Maxima based multi-scale edge information fusion method.Finally, a three-layer BP neural network model is constructed by pre-processing the grayscale variables in the column direction of magneto-optic images by PCA method, and obtaining 256 feature points which can represent the information of grayscale variables in the column direction of 95% magneto-optic images.The welding defect samples are classified.The experimental results show that the proposed method can accurately identify the welding defects such as micro-depression, non-fusion and welding deviation, and the accuracy of model classification can reach 90.80.
【作者单位】：广东工业大学机电工程学院广东省计算机集成制造重点实验室;广州番禺高勋染整设备制造有限公司;
【基金】：国家自然科学基金资助项目(No.51175095) 广东省协同创新与平台环境建设专项基金资助项目(No.2015B090901013) 广东省重大科技专项资助项目(No.2014B090921008) 广州市科学研究专项基金资助项目(No.201510010089) 佛山市科技创新专项基金资助项目(No.2014AG10015)
【分类号】：TP391.41

【相似文献】