基于SOM神经网络聚类以及支持向量机的数控机床热误差建模方法的研究∗

摘要

In order to reduce the impact of seasonal thermal error of spindle on CNC machine tool, a novel method of modeling based on preclassification by SOM neural network and Support Vector Regression Ma-chine(SVRM) is proposed. Experiment was performed on a HTM40100h turning milling machining center. Temperature and error data both in Summer and Winter were collected, and the measuring points of temperature on machine tool were classified into internal heat sources points and external ones. The temperature data of exter-nal heat sources points were input into SOM neural network for seasonal classification. The data of classified ex-ternal heat sources points along with the meanwhile data from internal heat sources points were input into corre-sponding SVRM model for error fitting. The comparison between modelling with and without preclassification shows that, the former one presents much better robustness and precision in two different seasons.%为了减小主轴季节性热误差影响，提高机床的加工精度，提出了基于针对机床热源进行SOM神经网络预聚类后的支持向量回归机的主轴热误差综合模型。针对一台型号为HTM40100 h的车铣复合中心，对主轴的关键温度测点进行了内外热源的划分，并在冬夏两个季节对所有测温点温度和热误差数据进行采集，将外部热源温度数据作为SOM网络的输入变量进行季节性聚类，聚类后的外部热源温度数据连同同时刻的内部热源温度数据一起作为不同季节支持向量回归机模型的输入变量，得到热误差拟合值。将通过聚类预处理的方法与未经聚类的方法进行了对比试验，结果表明：该综合预测模型在冬夏两个季节均获得了较高的建模精度和鲁棒性。