Abstract: To solve the problem of instable prediction of single BP neural network model,a thermal error modeling method for CNC machine tool spindle based on ensemble BP neural network was proposed.The Fuzzy c-means Clustering Method (FCM) was adopted to select the temperature sensitive points,which eliminated the multi-collinearity between redundant temperature variables.From the perspective of machine learning,several typical BP neural network models with weak prediction performance were integrated by average method,median method and ordinary least squares method respectively.By taking a horizontal machining center THM6380 as an example,the thermal experiment was carried out.The analysis results of thermal error prediction performance indicated that the prediction accuracy and generalization ability of the ensemble models were better than that of single BP neural network model.The proposed modeling method provided a new idea for the thermal error modeling and subsequent thermal error compensation of machine tool spindle.

Key words: spindle thermal error, BP neural network, fuzzy c-means clustering, ordinary least squares, ensemble model

摘要： 为了解决单一BP神经网络模型预测性能不稳定的问题,提出一种集成BP神经网络的数控机床主轴热误差建模方法。采用模糊c均值聚类法筛选温度敏感点,消除了冗余温度变量间的多重共线性。从机器学习的角度出发,分别采用平均法、中位数法和普通最小二乘法将几种具有弱预测性能的典型BP神经网络模型进行集成。以THM6380卧式加工中心为研究对象进行了主轴热误差实验,热误差预测性能分析结果表明,集成模型的预测精度和泛化能力优于单一BP神经网络模型,为机床主轴热误差建模及后续热误差补偿提供了新的思路。

关键词: 主轴热误差, BP神经网络, 模糊c均值聚类, 普通最小二乘法, 集成模型