Research on thermal error modeling of CNC machine tools based on PLS-BP neural network.

BP neural network is characterized by fitting nonlinear data in machine tool thermal error modeling, but it has poor robustness. This paper proposes a PLS-BP neural network modeling method to effectively improve the prediction accuracy and robustness of the model. First, this method employs the partial least squares to reduce the dimensionality of temperature data and extract principal components, eliminating redundant information. Second, a regression mapping model is established on the basis of BP neural network and thermal error, and its prediction performance is compared with that of traditional BP models. The results indicate the PLS-BP modeling method achieves high prediction accuracy and robustness. It is also shown that the method is markedly superior to the traditional BP models, keeping the maximum mean and maximum standard deviation of residual standard deviation sets of prediction results below 3.13 jxm and 1.32 jxm respectively [ABSTRACT FROM AUTHOR]