基于高清测量的三维零件表面滤波及分类

发布时间：2018-02-13 02:15

本文关键词： 高清测量表面形貌滤波分类　出处：《上海交通大学》2015年硕士论文　论文类型：学位论文

【摘要】：机加工零件表面形貌不仅直接影响零件的使用性能,还可以为加工过程提供反馈信息。因此很有必要对零件表面形貌进行准确的分析和评定。表面测量技术的快速发展尤其是高清测量技术的出现,使得快速地获取表面点云数据成为可能。本文正是基于高清测量技术,以三维零件表面形貌为研究对象,进行表面滤波和分类两个方面的研究。本文分析了已有的表面滤波方法,注意到小波滤波有着明显的优越性并且是当下的研究热点。然而已有的小波滤波方法主要针对两维表面,并不完全适用于三维表面的情形。基于此,本文将多尺度几何分析的最新成果—剪切波,引入到三维零件表面滤波领域。建立了剪切波滤波框架并对其表面计量特性进行了分析,通过仿真实验和多个实例分析对剪切波滤波进行了验证,并与国际标准ISO 11562规定的高斯滤波进行了比较。分析结果表明:1)剪切波具有多尺度分辨能力、高度的方向选择性、平移不变性和计算效率高等优点;2)剪切波具有零相移和理想的幅度传输特性;3)对于不同类型的工程表面,剪切波滤波都能很好地实现粗糙度、波纹度和形状误差的分离。4)分别根据剪切波滤波结果和高斯滤波结果计算的三维表面参数值的偏差在5%-25%之间,说明了剪切波滤波与作为国际标准的高斯滤波得到的结果是较为接近的。零件表面分类是识别零件表面质量的有效手段之一。本文提出了零件表面多类别分类的MPO-SVME算法。该算法以三维表面高清测量数据为输入,用支持向量机(SVM)构造基分类器,并通过选择性集成策略选择最优的分类器来提高分类正确率。通过发动机缸体顶面的实测三维数据验证了本文所提的分类方法,数值实验结果表明:使用单个SVM多类分类器对缸体顶面的分类正确率为89.58%,而MPO-SVME算法可以将平均分类正确率提高到92.22%,并且平均只选择2-3个基分类器。因此,MPO-SVME算法在只集成少量基分类器的情况下,能够有效地提高零件表面的分类正确率。
[Abstract]:The surface morphology of machined parts not only directly affects the performance of the parts, Therefore, it is necessary to analyze and evaluate the surface morphology of parts accurately. The rapid development of surface measurement technology, especially the appearance of high-definition measurement technology, can provide feedback information for machining process. It is possible to obtain the surface point cloud data quickly. In this paper, based on the high-definition measurement technology, the surface morphology of 3D parts is taken as the research object. Surface filtering and classification are studied in this paper. It is noted that wavelet filtering has obvious advantages and is a hot research topic at present. However, the existing wavelet filtering methods are mainly aimed at two-dimensional surfaces and are not fully applicable to the case of three-dimensional surfaces. In this paper, the latest achievement of multi-scale geometric analysis, shear wave, is introduced into the field of 3D surface filtering. The frame of shear wave filtering is established and its surface measurement characteristics are analyzed. The shear wave filtering is verified by simulation experiments and several examples, and compared with Gao Si filter according to the international standard ISO 11562. The analysis results show that the shear wave has multi-scale resolution and high directional selectivity. Translation invariance and high computational efficiency are advantages of shear wave with zero phase shift and ideal amplitude propagation characteristics. For different types of engineering surfaces, shear wave filtering can achieve good roughness. Separation of corrugation and shape error. 4) the deviation of 3D surface parameters calculated from the results of shear wave filtering and Gao Si filtering is between 5% and 25%, respectively. It is shown that the result of shear wave filtering is close to that obtained by Gao Si filter as an international standard. Surface classification of parts is one of the effective methods to identify the surface quality of parts. In this paper, a multi-class classification of parts surface is proposed. MPO-SVME algorithm. This algorithm uses 3D surface high-definition measurement data as input, Support vector machine (SVM) is used to construct the base classifier, and the optimal classifier is selected by the selective integration strategy to improve the classification accuracy. The proposed classification method is verified by the measured 3D data on the top surface of the engine cylinder block. The results of numerical experiments show that the classification accuracy of cylinder top surface is 89.58 using a single SVM multi-class classifier, while the average classification accuracy of MPO-SVME algorithm can be raised to 92.22 and only 2-3 base classifiers are selected on average. When only a small number of base classifiers are integrated, It can effectively improve the classification accuracy of parts surface.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TH161.1

【相似文献】