【作者】 徐金忠；

【导师】 叶文华；

【作者基本信息】 南京航空航天大学 ， 机械电子工程， 2009， 硕士

【摘要】 论文在分析国内外数控机床热误差补偿技术研究现状及发展趋势的基础上,以数控机床热误差补偿中温度与热误差检测系统的开发、热关键点辨识技术、热误差建模方法为主要研究内容。开发了温度与热误差检测系统。系统硬件以温度传感器、位移传感器和PCI数据采集卡为核心;数据连续采集模块采用局部循环缓存技术,提高了整个采集程序的运行速率。为实现较高的测量精度,采取了包括信号调理电路在内的一系列防干扰措施。经试验,此采集系统满足测量精度和自动采集的要求。在机床热关键点的辨识研究中,针对聚类分组法当温度变量组之间出现交叉现象时分组难以进行的问题,提出了改进方法,即典型温度变量选择方法。该方法对温度变量与热误差观测值之间相关性高的变量优先分析,选出典型温度变量;用多元线性回归模型对各变量组合的模型拟合精度进行分析,从而优选出温度变量组合。研究了基于最小二乘的多元线性回归、BP神经网络和RBF神经网络三种热误差建模方法。采用这三种建模方法对HG-410立式数控铣床和MCH63卧式加工中心分别建立其热误差模型,通过试验验证了各模型的热误差预测能力。分析比较各模型的热误差预测精度,综合考虑模型的复杂程度,得出多元线性回归法适用于HG-410立式数控铣床进行热误差建模,BP神经网络适用于MCH63卧式加工中心进行热误差建模的结论。更多还原

【Abstract】 Based on fully understanding and deeply analyzing the current status of the research and trend of thermal error compensation technique for CNC machine tools, the development of temperature and thermal error data acquisition system, identification technique of thermal key points and the modeling methods of thermal errors for CNC machine tools were the main research contents of this paper.A temperature and thermal errors measurement system was developed. It’s hardware was based on temperature sensors, displacement sensors and PCI data acquisition card; the whole operation rate of data acquisition program was enhanced by using local circulation cache technology for dada continuous acquisition module. In order to acquire high measuring accuracy, a series of measures including signal processing circuit were adopted. Test results indicated that the system completely met the requirements of precision and automatic measurement.In the research of thermal key points identification, in order to solve the problem of when the crossover phenomenon appeared among temperature variable groups grouping was hard to implement, a improved method ,selection method of typical temperature variables was proposed; in order to choose the reasonable temperature variable combination multiple linear regression (MLR) model was also used.The modeling methods of multiple linear regression based on least square method,BP neural network and RBF neural network were studied. Taking HG-410 vertical CNC machine tools and MCH63 horizontal machining center as the test objects, the models of these machine tools were established respectively by the three modeling methods, and the prediction ability of these models were proved by experiments. By comparison of the prediction accuracy and comprehensive consideration of the complex degree of these models , a conclusion is: MLR modeling methods is suitable for HG-410 vertical CNC machine tools and BP neural network is suitable for MCH63 horizontal machining center.更多还原