An Experiment-Based Multi-Objective Modeling Method for Thermal Errors of Slant Bed CNC Lathes
Yan Zongzhuo,Tao,Du Hongyang,Shi Hu,Mei Xuesong
DOI: https://doi.org/10.1007/s00170-022-09158-4
2022-01-01
Abstract:Due to the special design of slant bed CNC lathes, the temperature of the X-axis feed motor will significantly increase after power-up, even if there is no feed motion. Based on this phenomenon, we redesigned the experiments that are used to study the effects of ambient temperature variations, the heat generated in the spindle, and the heat generated in the X-axis feed system on the thermal errors. The measurement scheme was improved to identify the effects of the tilt of the spindle axis and turret. The heat generated in the X-axis feed system was divided into two categories: the heat unrelated and related to the feed motion, based on their sources, and their effects were studied separately in different experiments. Since the relationship between the temperature distribution and thermal errors was relatively explicit in each newly designed experiment, the effects of each major factor could be accurately and reasonably described by the established thermal error models. According to our analysis, the conclusions and models obtained in the newly designed experiments can provide a reliable reference for thermal error modeling under complicated operating conditions. Based on these findings, a new experiment-based multi-objective modeling method implemented with the multi-objective particle swarm optimization (MOPSO) algorithm was proposed. In addition to the prediction accuracy, the reasonability of the model coefficients, which was quantified by the difference between the candidate and reference values, was identified as an important optimization objective. Compared with traditional methods, the models established with the proposed method expressed the relationship between the temperature and thermal error more reasonably, and it could successfully maintain its accuracy at different spindle speeds. According to the experimental verification, the residual errors were at least 55.28% lower when predicting the thermal error in the X-direction with the proposed method.