基于机器视觉的列车外齿轮磨损状态检测方法

Defect Status Detection Method Based on Machine Vision for External Gear of Train

齿轮是高铁列车中的重要部件,齿轮磨损状态的程度对列车运行安全具有重要的影响。针对采用人工观察确定齿轮磨损状态的问题,提出一种基于机器视觉的列车外齿轮磨损状态定量检测方法。为避免将顶部和底部低灰度啮合区分割为背景,提出分块分割算法以得到候选啮合区域。为去除候选啮合区中的背景区域且将啮合区分割为一个整体,提出区域聚合算法。为避免点蚀区域位于啮合区边缘造成的不完整分割问题,提出基于凸包运算的边缘修正算法。结合分块分割、区域聚合以及边缘修正,实现齿面图像啮合区分割。结合自适应局部阈值以及基于形状特性的假阳性去除算法,实现齿面图像的点蚀区域检测。在140幅齿面图像上对提出方法进行验证,啮合区分割的平均AOM为0.89,点蚀区域检测方法性能优于现有方法。

As Gear is an important component of a high-speed train,the defect status of the gear affects train operation safety.To solve the problem of defect status measurement through visual inspection,a quantitative defect status detection method based on machine vision for the train external gear was proposed.To avoid segmenting the low-gray top or bottom meshing region as background,a sectional segmentation algorithm was proposed to obtain the candidate meshing region.To eliminate the background region in the candidate meshing region and merge the meshing regions as a whole,a region merge algorithm was presented.To avoid the incomplete segmentation caused by the defect locating on the periphery of the meshing region,an edge refinement algorithm based on convex hull operation was designed.Combined sectional segmentation,region merge and edge refinement,the meshing region segmentation in the gear image was implemented.The detection of the surface defect in the gear image was realized,using local adaptive thresholding and false positive elimination algorithm based on shape characteristics.The proposed method was tested on 140 gear tooth images.The result for meshing region segmentation is 0.89 under the area of overlap measurement(AOM).The proposed defect detection method shows better performance than some existing related approaches.