摘要
针对机床热误差补偿技术中预测模型建立的问题,综合多元线性回归及BP神经网络的优点,提出一种机床热误差建模新方法。由不同样本数据建立若干多元线性回归模型,依据统计学理论筛选出预测精度及鲁棒性高的回归模型,预处理后将其结果输入到BP神经网络中进行非线性拟合建模,在不断调节网络权值及对神经网络训练的基础上,最终建立热误差补偿模型。在卧式加工中心上进行试验验证,主轴Z向最大热误差从17.895μm减小到1.654μm。
A new method is proposed to build the thermal error compensation model considering the advantages of multiple linear regression and BP neural network. Some multiple linear regression models are established by using different sample data. According to the statistics principle, some high accurate and robust models are selected. The predicted results are input into BP neural network for nonlinear fitting modeling. The final thermal error compensation model is built on the basis of adjusting the network weights and training the neural network constantly. The method is validated on a horizontal machining center. The result shows that Z-axis thermal error is reduced from 17. 895μm to 1. 654μm.
出处
《组合机床与自动化加工技术》
北大核心
2016年第1期63-66,70,共5页
Modular Machine Tool & Automatic Manufacturing Technique
基金
大连市科技计划项目"汽车发动机和柔性自动线可靠性技术研究"(2013A11GX012)
国家"高档数控机床与基础制造装备"科技重大专项课题(2013ZX04012071)
关键词
热误差模型
多元线性回归
统计学
BP神经网络
thermal error model
multiple linear regression
statistics principle
BP neural network
