精密薄零件微米级轮廓测量系统的分析与设计

发布时间：2018-04-14 12:26

本文选题：精密薄零件 + 边缘检测　；参考：《电子科技大学》2016年硕士论文

【摘要】：精密薄零件是一类尺度较小,但在工业领域有重要应用的零件。这类精密零件的边缘形态对其工作特性有重要影响,因此此类精密薄零件的边缘是检测的重要指标。但由于这类零件的尺度小,且边缘薄,常规检测手段难以进行有效检测。摄影测量作为一种现代工业检测的重要手段,是解决这类精密零件检测问题的有效方式。摄影测量的核心是数字图像处理技术,图像处理技术在边缘检测领域已有诸多研究。但当图像检测的精度要求与像素尺度数的量级相近时,即使是单个像素的检测差距都会对最终的检测结果有较大的影响。而传统的图像处理算法在输出最后结果时都会执行“去模糊”操作,即将一个信息不确定的像素映射到一个非黑即白的二元集合,在小尺度下,图像的边缘信息往往是模糊的不确定的。这样的去模糊操作会对精密薄零件的边缘检测造成精度的损失。基于这个考虑,本文提出了基于模糊集的图像边缘检测算法,与传统方法不同的是,本文将模糊集内的每个元素都可能是图像的边缘,引入全向梯度算子的概念,以全向梯度算子来衡量像素属于图像边缘的可能性。同时,不同于传统方法执行的去模糊操作,本文算法将模糊集中的模糊信息一直保留,最终得到的边缘不是一个边缘点集,而是一个边缘模糊集。并且,为了减小模糊集内元素的个数,剔除明显不合适的元素,提出基于像素信号能量的ROI提取算法:将像素块的块内方差建模为信号能量,能量越大的区域其属于边缘的可能性越大,设定阈值,剔除那些在总的信号能量中占比极小的区域。在这两步的基础上,文章独立推导了模糊集上的最小二乘圆心拟合算法,得到了模糊集下的圆心公式。由圆心公式可进一步求得模糊集上各元素转化到极坐标下,并与CAD数据进行对比,在一定显著性水平下,即可找出误差异常值完成对零件图像的检测。经求解,可以正确地检测出示例样本中第69号齿尖(在极角347.3229,352.0326,348.7766,350.7192,351.3807,347.6453,347.3832)处存在明显异常。经人工复检,本算法的漏检率达0%,误检率为60%,可以有效的检测出精密薄零件的异常点。
[Abstract]:Precision thin parts are a kind of small scale parts, but have important applications in the industrial field.The edge shape of this kind of precision part has an important influence on its working characteristics, so the edge of this kind of precision thin part is an important index of detection.However, because of its small scale and thin edge, conventional detection methods are difficult to detect effectively.As an important means of modern industrial detection, photogrammetry is an effective way to solve the problem of precision parts detection.The core of photogrammetry is digital image processing, which has been studied in the field of edge detection.However, when the accuracy of image detection is similar to the magnitude of pixel size, even the detection gap of a single pixel will have a great impact on the final detection results.Traditional image processing algorithms perform "de-blur" operations when they output final results, that is, a pixel with uncertain information is mapped to a non-black or white binary set, and at a small scale,The edge information of an image is often fuzzy and uncertain.Such a deblurring operation will result in a loss of accuracy in the edge detection of thin precision parts.Based on this consideration, an image edge detection algorithm based on fuzzy sets is proposed in this paper. Different from traditional methods, every element in a fuzzy set is likely to be the edge of an image, and the concept of omnidirectional gradient operator is introduced in this paper.The omnidirectional gradient operator is used to measure the possibility that pixels belong to the edge of the image.At the same time, different from the traditional de-fuzzy operation, the fuzzy information in the fuzzy set is always preserved by the algorithm, and the resulting edge is not a set of edge points, but a set of edge fuzzy.Furthermore, in order to reduce the number of elements in the fuzzy set and eliminate the obviously inappropriate elements, a ROI extraction algorithm based on pixel signal energy is proposed: the intra-block variance of the pixel block is modeled as the signal energy.The larger the energy is, the more likely it is to belong to the edge. The threshold is set to eliminate the regions that account for a very small proportion of the total signal energy.On the basis of these two steps, the least square centroid fitting algorithm on fuzzy sets is derived independently, and the center formula for fuzzy sets is obtained.The elements in the fuzzy set can be further transformed to polar coordinates by the center of circle formula, and compared with the CAD data. At a certain level of significance, the abnormal value of the error can be found to complete the detection of the part image.After solving, we can correctly detect the obvious anomaly at the tip of tooth No. 69 in the sample (347.3229352.0326) at the polar angle 348.7766350.7192v 351.3807347.6453n 347.3832.After manual rechecking, the missed detection rate of this algorithm is 0 and the false detection rate is 60. It can effectively detect the abnormal points of the precision thin parts.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TH71;TP391.41

【参考文献】