科研之友
微信
新浪微博
Facebook
分享链接
摘要
Thermal error significantly influences the accuracy of CNC machine tool. A high-performance compensation system depends upon the accuracy and robustness of the model and appropriate model inputs of temperatures. In this paper, fuzzy clustering is conducted in temperature classification. Based upon an evaluation model, an optimal temperature classification can be found. The representative temperatures from each group construct temperature candidates. Then, a sliced inverse regression (SIR) model is introduced in thermal error modeling, which can change the problem of high-dimensional forward regression into several one-dimensional regression and meanwhile further eliminate the coupling among temperatures candidates. To evaluate the performance of SIR model, measuring experiment was carried out on a horizontal machining center for temperature and thermal error information under two experimental conditions. The proposed classification method was used to classify 29 temperature variables into five groups. A SIR model was built upon the five temperature candidates. Meanwhile, stepwise regression (SR) theory is also conducted for temperature variable selection and thermal error modeling. Comparison shows that both of the two models have high fitting accuracy, while the SIR model is more excellent in robustness than the SR model. Finally, a compensation system was developed. Compensation experiment shows that the SIR model is practical and effective, which can reduce the axial thermal error from 43 to 7 mu m.
