Abstract: In order to improve the machine tool accuracy,a new modeling method of CNC machine tool thermal error which is based on the Kalman Filter is presented.Regard the regression coefficient of statistic model as a state vector,take the statistic model for the observation equation,and the statistic model has been realized by using the Kalman Filter method.Compared with the least squares,the Kalman method which is the least linear estimator of variance can be obtained the higher accuracy.The changing temperature of the spindle and the heat deformation is measured at the same time by using the transducer on the CNC machine tool.The model of Kalman filter is compared with the model of the least squares and the LS-SVM method.It shows that the accuracy of Kalman filter model is 10.5 % and 1.8 % higher than the LS and LS-SVM model.Moreover,the modeling time of Kalman method is 0.9 % and 6.8 % less than the modeling time of LS and LS-SVM method.

Key words: Kalman filter, thermal error, CNC machine tool spindle, model