基于主成分分析与BP神经网络相结合的机床主轴热漂移误差建模

Thermo-Drifting Error Modeling of Spindle Based on Combination of Principal Component Analysis and BP Neural Network

为提高数控机床热误差模型的预测精度,提出了将主成分分析与BP神经网络相结合的主轴热漂移误差的建模和预测方法.使用主成分分析法对多个温度变量进行降维处理或重新组合,将处理后所得较少的主成分变量作为样本输入BP神经网络进行训练而得到主轴热漂移误差模型,并与经过测点优化后以关键点温度作为输入的BP神经网络模型进行对比分析.结果表明:基于主成分分析与BP神经网络相结合的主轴热漂移误差模型的拟合精度较高,残差较小;由于BP神经网络的输入变量较少而使所提出的模型训练速度快、迭代次数少.

In order to improve the prediction accuracy of thermal error model, a data processing method based on the combination of principal component analysis and BP neural network was presented. By using principal component analysis, the amount of temperature variables will be reduced. Then the principal components are employed to train the BP neural network in order to obtain the thermal error model. As the number of inputs is reduced, the train process can be faster and the iteration time can be reduced. In comparison with the network model which uses the critical point temperatures as the input, the results show that BP neural network thermal error modeling method based on principal component analysis meth- od has advantages of high fitting accuracy and smaller residual error. So this modeling method has both high efficiency and accuracy of compensation.