Selecting The Temperature Measurement Points Of The NC Machine Tool And Constructing The Compensated Model Of The Thermal Error

Thermal error is the biggest error source of processing precision of NC machine tool. It can be 70 percent of the total error. To improve the processing precision of NC machine tool, the thermal error should be compensated effectively. The key issues which need to be solved are how to select the best temperature measurement points and how to construct the effective model for compensating the thermal error. The study works of this thesis are to solve the two key issues.Aiming at selecting the best temperature measurement points, the new method, which combines the finite element and the Kohonen neural network, is proposed. Firstly, in order to figure the temperature of the analytical object, the ANSYS is used to construct the finite element entity model. Then the coupling of heat and structure is used to compute the thermal deformation quantity of the analytical object. Secondly, the node temperature and the thermal deformation quantity of the spindle computed in the ANSYS, which are considered as the data sample. Then the Kohonen neural network is used to select the points with the data sample. The result indicates that the process of selecting the measurement points based on this new method is implemented easily. To show the feasibility and validity, this new method is used to analyze the CK6132 NC lathe. Based on the theory and the new method, the temperature sensors on the NC machine tool can be arranged effectively.To construct the effective model for compensating the thermal error of NC machine tool, the Radial Basis Function neural network based on the Kohonen compete rule is proposed. The Kohonen compete rule is used to obtain the center vectors of the latent layer. The result indicates that the arithmetic is simple and availability. The less operand and the shorter runtime are held in this method. The capability of RBF neural network is enhanced. And then this method is used to modeling for compensating the thermal error of NC machine tool. The predict results of this model are compared with the result of GRNN neural network, the result of the BP neural network and the result of the Multivariable linear regression. The two advantages of the model of the RBF neural network based on the Kohonen learning rule are shown. The higher predict precision and the fast runtime of the procedure are indicated. So the method can be used suitably in the real time compensating control of the thermal error on the NC machine tool. Lastly, to realize the predictable function of the neural network conveniently, the Active technology is used to transfer the MATLAB application into the VB procedure.