基于图形识别的数控机床运动误差快速溯因与推算

发布时间：2018-05-11 14:08

本文选题：数控机床 + 运动误差　；参考：《重庆理工大学》2015年硕士论文

【摘要】：运动误差,作为数控机床误差的最终反应,包含了数控机床的几何误差和控制误差的信息,对数控机床的加工精度有着重大的影响;如何快速准确的监控生产线上运行的数控机床的运动误差,对抑制加工产品批量事故的发生及提高企业的生产效率具有积极的意义。而当前学者们对于数控机床精度的研究偏重于检测、控制与补偿等领域,对误差溯源领域的的研究却较少。已有对误差溯源的研究多采用误差建模,方法复杂且依赖数控机床的结构和类型,适用范围较小。本文提出利用数控机床圆运动轨迹的图形,采用图形识别相关技术,定义一种新的特征角点并开发出角点检测算子,检测此角点在圆运动轨迹图形上的分布规律;将圆周分割为16维,分析各维上可反映该维整体特征的平均半径和反映局部特征的角点个数,从而建立可反映图形特征的三维特征矩阵,并采用支持向量机对特征矩阵到误差图形的映射的鲁棒性做了验证,最后结合径向基函数神经网络实现运动误差源溯因网络的构建,实验结果显示该方法识别准确率高,识别速度快,简便而高效。更开发了软件系统,简洁明了的界面,具有优良的友好性。采用本文所述方法尤其适合机床使用企业在制造过程中的精度溯因与控制,方法经济简便,具有较大的实际应用价值。本文的主要研究内容分为图形的角点检测、图形特征提取、特征矩阵到误差图形映射鲁棒性的验证和综合误差溯源网络的建立三个部分:首先是图形的识别工作,本文提出一种新的特征角点,采用角点检测的方法,将采集来的圆运动误差轨迹图形经过预处理之后再分割为16维,经过文中设计开发的角点检测器检测出符合定义的特征角点。其次,研究特征角点在分割为16维的误差轨迹图形的分布规律,计算各维圆周上所有点的平均半径和特征角点个数,构建了一个三维特征矩阵,从而建立了误差图形与特征矩阵的映射关系。并采用支持向量机对该映射关系进行了验证,结果显示支持向量机对实验样本的分类效果显著,表明文中所建立的映射关系鲁棒性强。最后,建立基于径向基函数神经网络的综合误差溯因网络,以特征矩阵为输入,各单项误差源为网络输出。经过训练的溯因网络最终实现了对综合误差的快速溯因,且识别率较高。
[Abstract]:Motion error, as the final reaction of NC machine tool error, contains the information of geometric error and control error of NC machine tool, which has a great influence on the machining accuracy of NC machine tool. How to quickly and accurately monitor the movement error of NC machine tools running on the production line is of positive significance to restrain the occurrence of batch accidents and to improve the production efficiency of enterprises. At present, the research on the accuracy of NC machine tools is focused on the fields of detection, control and compensation, but the research on error traceability is less. Error modeling is often used in the research of error traceability. The method is complex and depends on the structure and type of NC machine tools, and the scope of application is relatively small. In this paper, a new characteristic corner is defined and a corner detection operator is developed to detect the distribution of the corner on the circular motion trajectory by using the graph of the circular motion trajectory of the NC machine tool and the correlation technology of the graph recognition. The circle is divided into 16 dimensions, and the mean radius of the whole feature and the number of corner points reflecting the local feature on each dimension are analyzed, and the 3D feature matrix which can reflect the feature of the graph is established. The robustness of the mapping from feature matrix to error graph is verified by support vector machine. Finally, the radial basis function neural network is used to construct the moving error source tracing network. The experimental results show that the method has high recognition accuracy. The recognition speed is fast, simple and efficient. More developed software system, simple and clear interface, with good friendliness. The method described in this paper is especially suitable for the precision tracing and control in the manufacturing process of the machine tool enterprises. The method is economical and simple and has great practical application value. The main content of this paper is divided into three parts: corner detection, feature extraction, robustness verification from feature matrix to error graph mapping and establishment of comprehensive error traceability network. In this paper, a new feature corner is proposed. By using corner detection method, the collected circular motion error trajectory is preprocessed and then divided into 16 dimensions. The corner detector designed in this paper detects the characteristic corner in accordance with the definition. Secondly, the distribution law of the characteristic corner in the 16 dimensional error trajectory pattern is studied, and the mean radius and the number of characteristic corner points of all points on each dimensional circle are calculated, and a three-dimensional feature matrix is constructed. The mapping relationship between the error graph and the feature matrix is established. Support vector machine (SVM) is used to verify the mapping relationship. The results show that the classification effect of SVM on experimental samples is significant and the mapping relationship established in this paper is robust. Finally, a comprehensive error traceability network based on radial basis function neural network is established. The characteristic matrix is used as input and the single error source is network output. The trained backtracking network finally realizes the fast traceability of the synthetic error, and the recognition rate is high.
【学位授予单位】：重庆理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41;TG659

【参考文献】