基于改进ResNet-18模型的弧齿锥齿轮安装位置偏差分类识别方法研究
Research on classification and identification method of installation position deviation of spiral bevel gears based on improved ResNet-18 model
刘小凯 1边骥轩 2张江勇 1董岑鑫 1王卫军1
作者信息
- 1. 广州先进技术研究所 机器人与智能装备中心, 广东 广州 511458
- 2. 中国北方车辆研究所,北京 100072
- 折叠
摘要
针对现有的弧齿锥齿轮安装位置偏差分类识别效率低的问题,使用改进的ResNet-18 神经网络构建轻量级弧齿锥齿轮安装位置偏差分类识别模型,改进的网络修改了ResNet-18网络第一卷积层,并引入了通道与空间注意力机制.实验结果表明,改进ResNet-18 较原网络对于齿面接触印痕图像的分类准确率提高了 3.23%,损失值降低了 0.04.该方法为齿轮安装偏差识别提供了新路径,对弧齿锥齿轮总成的安装调整作业提供了一定的指导.
Abstract
Aiming at the problems of low efficiency in identification of spiral bevel gear installation position deviation,the improved ResNet-18 neural network was used to build a light-weight spiral bevel gear installation position deviation classification model.The improved network modified the first convolutional layer of the ResNet-18 network,and introduced a channel and spatial attention mechanism.The experimental results showed that,for gear contact pattern identification,the accuracy of the improved ResNet-18 was 3.23%higher than that of the original network,and the loss value was reduced by 0.04.The proposed method offers a new path for gear installation deviation identification,and provides guidance in assembling the spiral bevel gear.
关键词
弧齿锥齿轮/接触印痕/安装位置偏差/注意力机制/ResNet-18Key words
spiral bevel gear/contact pattern/position deviation/attention mechanism/ResNet-18引用本文复制引用
基金项目
国家重点研发计划项目子课题(2018YFA0902901)
出版年
2024
国家科技期刊平台
NSTL
万方数据