Thermal error prediction of machine tool spindle using segment fusion LSSVM

Abstract

The key temperature points are the input variables of the thermal error model for prediction and compensation of thermal errors for precision CNC machine tools. However, the revealed time-varying characteristics of the key temperature points may jeopardize the robust prediction. To this end, the segment fusion least squares support vector machine (SF-LSSVM) thermal error modeling method is proposed. Firstly, the temperature data and thermal error data are divided into different segments according to time. Then, using the LSSVM with excellent nonlinear mapping capabilities as the basic model, the sub LSSVM thermal error model building and the corresponding key temperature points selection in each segment are fulfilled with genetic algorithm (GA) in a wrapper manner to preserve the corresponding local prediction characteristics. Finally, pick some of or all the sub LSSVM thermal error models to fuse together as the final thermal error model which may incorporate both the local and global prediction characteristics. The modeling and prediction experiment results on the spindle thermal error of a horizontal machining center demonstrate that the mean root-mean-square error (RMSE) on 5 spindle speeds after compensation is only 3.1 μm. Comparing with two traditional thermal error models, the prediction performance of the present model is improved by up to 51%. This research casts new light on both the mechanism of key temperature points and the prediction method of thermal errors.