摘要
针对现有的热误差建模方法建模效率低,模型预测精度不理想等问题,提出了广义径向基函数神经网络(RBF)建模方法并将其应用于数控机床热误差建模中。讨论了采用广义RBF神经网络进行热误差建模的原理及步骤。以数控导轨磨床主轴箱系统为例,布置了12个主轴热误差的关键温度测点,测得了2组独立的主轴箱系统热误差数据。将测得的数据分别用于建立主轴箱系统热误差广义RBF神经网络预报模型和验证模型的准确性。研究结果表明,热误差广义RBF神经网络模型具有预测精度高及泛化能力强的优点;与传统的RBF神经网络建模方法相比,提出的广义RBF神经网络建模方法建模效率更高,模型鲁棒性及预测性能更好,是一种可以用于数控机床热误差实时补偿的有效建模方法。
In considering of the lower efficiency and worse prediction accuracy of the existing thermal ／ error modeling methods, a generalized Radial Basis Function （RBF） neural network modeling approach was proposed to establish the thermal error model of Numerical Control（NC） machine tools. The model theory and the corresponding steps based on this method were discussed. An experiment on the spindle box of a NC guide rail grinder was performed, and two groups of independent thermal error data were obtained by setting twelve critical temperature measuring points of spindle thermal error. One group of the data was used for building the thermal error model of the spindle box system based on generalized RBF neural network method and the other was used for verifying the correction of the model. The study results show that the thermal error model based on generalized RBF neural network method has high prediction precision and good generalization ability. By comparing the generalized RBF neural network method with traditional RBF neural network modeling method, the former shows better efficiency, robustness and prediction capacity, and it is an effective modeling method for the real-time thermal error compensation of NC machine tools.
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2015年第6期1705-1713,共9页
Optics and Precision Engineering
基金
国家自然科学基金资助项目(No.51175161
No.51475152)
国家科技重大专项资助项目(No.2011ZX04003-011)
关键词
广义径向基函数
神经网络
热误差建模
聚类算法
泛化能力
鲁棒性
数控导轨磨床
generalized radial basis function
neural network
thermal error modeling
clusteringalgorithm
generalization ability
robustness
Numerical Control（NC） guide rail grinder
