铁路货车铸件DR图像的疏松缺陷快速检测  

Rapid detection of porosity defect in DR image of railway freight car casting

作　　者：李小利 曾理[1,2] LI Xiaoli;ZENG Li(School of Mathematics and Statistics,Chongqing University,Chongqing 401331,China;Engineering Research Center of Industrial Computed Tomography Nondestructive Testing of the Education Ministry of China,Chongqing University,Chongqing 400044,China)

机构地区：[1]重庆大学数学与统计学院,重庆401331 [2]重庆大学工业CT无损检测教育部工程研究中心,重庆400044

出　　处：《铁道科学与工程学报》2022年第6期1763-1773,共11页Journal of Railway Science and Engineering

基　　金：国家自然科学基金资助项目(61771003)。

摘　　要：快速有效地检测铁路货车铸件的疏松缺陷,对铁路运输安全具有重要意义。数字化X射线照相可以直接获取含有缺陷的铸件图像,成像速度快。但人工目测图像检测法不仅耗时费力容易疲劳,而且依赖于检测人员的经验。研究自动化检测方法实现图像缺陷的实时检测具有重要的意义。铸件图像中疏松缺陷形状复杂,可以分为树枝状、羽毛状和海绵状等,也可能是多种不同形状疏松缺陷的组合,这增加了检测识别的难度。发展基于深度学习的铸件数字化射线照相的疏松缺陷检测和识别方法。首先,对铸件数字化射线照相的原始图像进行一系列的预处理,采用引导图像滤波结合分数阶微分对图像进行增强,利用图像标注软件标注增强后的图像,得到标注数据集;然后,将标注数据和含有不同等级的疏松缺陷图像输入到YOLACT深度学习网络中,进行训练和测试。测试结果表明:该方法对单幅图像的平均检测时间为1.48 s,疏松缺陷的平均检测率为60.56%。该方法可快速检测出铸件疏松缺陷的等级和类别,为实时检测工业铸件缺陷提供一种辅助解决方法。Detecting the porosity defects of railway truck castings rapidly and effectively is of great significance to railway transportation safety.The images of casting with defects can be obtained directly from digital X-ray photography with high speed.However,the visual image detection method is time-consuming and exhausting,and highly depends on the experiences of inspectors.It is thus of great significance to study the automatic detection method for achieving to the real-time detection of image defects.However,the porosity defects in casting images are of complex shape,which can be divided into,for instance,dendritic,feathery and spongy types;the combination of porosity defects with different shapes is also commonly found,which increases the difficulty of defect detection and identification.A deep-learning-based method for detecting and identifying porosity defects from digital radiography images of castings was developed.First,a series of pre-treatments were carried out on the original digital radiographic images of castings.Guided image filtering and fractional differential were used to enhance the images,and the image labeling software program was used to annotate the enhanced images to obtain the annotation data set.Then,annotation data and porosity defect images with different grades were input into the YOLACT deep learning network for training and testing.The test results showed that the average detection time of single image was 1.48 s,and the average detection rate of porosity defects was 60.56%.The proposed method can quickly detect the grade and category of porosity defects in castings,thus providing a supplementary method for real-time detection of industrial casting defects.

关 键 词：铁路货车铸件 数字化射线照相 缺陷检测 图像处理 深度学习 

分 类 号：TH878[机械工程—仪器科学与技术] TP391[机械工程—精密仪器及机械]