A Study On Thermal Distortion Error Model Of CL-20A CNC Lathe

CNC machine tools are widely used in modern manufacturing. With thedevelopment of science and technology, the higher accuracy of CNC machine toolis needed. Theoretical and practical research has shown that the thermal error is oneof the main obstacles and constraints the improvement of CNC accuracy. With theproject support from Ministry of Science and technology of China, CL-20AHorizontal Lathe of Dalian Machine tool Factory is chosen as research target, therelevant theoretical and experimental study has been expanded, here, the topicmainly focus on building of machine tool thermal error, which will be adopted inthe online real-time machine tool error compensation.First, many temperature sensors are pasted on the machine tool based on heatsource characteristics and distribution of machine tools, three inductance sensorsare employed to detect thermal deviation from spindle to tool position in threedirections. Then, the experimental data will be analyzed by successive regressionanalysis to identify the temperature-sensitive point and its location. With thetemperature-sensitive point, further experiment is designed and executed to find therelationship between the temperature variation of the machine sensitive point andthe thermal deviation of the spindle.The large amount obtained experimental data is analyzed by support vectormachine modeling and regression analysis. At the same time, the BP neural networkand least squares method are tested also to provide reference for Support VectorMachine method. The modeling results show that support vector machine is moresuitable for predicting the machine tool error, BP neural network and the leastsquares method has lower forecasting accuracy. Furthermore, compared to theblack-box operation and not fixed predict results of the Neural Network method,Support Vector Machine method can get the clear formula for calculation, and thetheory process is very rigorous, which will be convenient and useful to program andimplant into the NC system to achieve real-time thermal error compensationcalculation. Compared to the Least Squares method, Support Vector Machinemethod removed the redundant data, only a part of support vector used for modeling, so the generalization ability is strong.