以热图像为输人的基于ResNet的机床主轴热误差建模方法

打开文本图片集

中图分类号：TG659；TP183DOI：10.3969/j.issn.1004-132X.2025.09.018

Thermal Image Input-based ResNet Method for Thermal Error Modeling of Machine Tool Spindles

LI Mingfan 1，2 YANG Long³4LI ShengGUO Huan°FU Guoqiang？，4*

1.College of Mechanical Engineering， Zhejiang University of Water Resources and Electric Power， Hangzhou，310018

2.Key Laboratory of Key Technologies for Mechanical Industry Hydroelectric Power Generation Pump Turbine，Zhejiang University of Water Resources and Electric Power，Hangzhou，310018 3.School of Mechanical Engineering，Southwest Jiaotong University，Chengdu，610031 4.Intelligent Manufacturing Dragon City Laboratory，Changzhou， Jiangsu，213000 5.Linker Technology Research Co.，Ltd.，Hangzhou，310000 6.Henan Yuguan Safety Development Co.，Ltd.， Zhengzhou，450000

Abstract：To achieve a high-precision and highly generalizable thermal eror model of machine tools ， a thermal image input-based ResNet method was proposed for thermal error modeling of CNC machine tool spindles.A thermal image dataset labelled was constructed with thermal eror rounding，and a ResNetbased classification model was trained for thermal error prediction using thermal images as inputs. Considering the regression characteristics of the machine tool thermal error time series，a regression output layer was constructed by integrating the probabilities of diferent classification labels from the clasification output layer in a weighted manner，enabling thermal error regresson prediction without retraining. The deep features of thermal images and the classification performance of the ResNet model were visualized，confirming the effctiveness of ResNet in feature extraction and strong clasification ability.Finally，the ResNet model was compared with GoogLeNet and VGGNet models under different operating conditions，demonstrating the high accuracy and generalization of the ResNet-based thermal error clasification and regression models. Key words：thermal image；spindle thermal error；ResNet classification model；regression prediction；feature visualization

0 引言

随着航空航天、模具加工等行业的快速发展，对复杂工件的精密制造需求逐渐增加，数控机床的精度也越来越重要[1]。（剩余15415字）