文献类型：期刊文献

中文题名：基于迁移学习和特征重用的MIM齿轮缺陷检测

英文题名：MIM gear defect detection based on transfer learning and feature reuse

作者：赵桐[1];雷保珍[1,2];王训伟[2];齐广浩[1]

第一作者：赵桐

机构：[1]北京联合大学北京市信息服务工程重点实验室,北京100101;[2]北京联合大学北京市智能机械创新设计服务工程技术研究中心,北京100101

第一机构：北京联合大学北京市信息服务工程重点实验室

年份：2021

卷号：40

期号：10

起止页码：129-131

中文期刊名：传感器与微系统

外文期刊名：Transducer and Microsystem Technologies

收录：CSTPCD;;北大核心:【北大核心2020】；CSCD:【CSCD_E2021_2022】；

基金：国家自然科学基金面上资助项目(5177050098);北京市朝阳区协同创新项目(CYXC1905)。

语种：中文

中文关键词：迁移学习;特征重用;卷积神经网络;缺陷检测

外文关键词：transfer learning;feature reuse;convolutional neural network(CNN);defect detection

摘要：针对金属注射成型(MIM)工业生产零件由于缺陷细微导致的肉眼识别率低和人力资源耗费严重的问题,设计了一种基于卷积神经网络(CNN)的缺陷检测识别方法。利用迁移学习解决当前由于训练数据少而导致识别率低的问题,并提出了一种轻量化特征重用网络模型,对MIM工艺生产的小模数齿轮进行缺陷检测。实验结果表明:训练模型在MIM齿轮缺陷检测的准确率达到98%以上,可对零件实现快速、准确的检测,该检测识别技术可以应用于同类的零件制造业完成分类和检测任务,在工业上具有重要的研究价值和实践意义。
Aiming at the problems of low visual recognition rate and serious human resource consumption of metal injection molding(MIM)industrial production parts due to the fine defects,a defect detection and recognition method based on convolution neural network(CNN)is designed.The problem of low recognition rate caused by the lack of training data is solved by using transfer learning.A lightweight feature reuse network model is proposed to detect the defects of small module gears produced by MIM process.The experimental results show that the accuracy rate of the training model in MIM gear defect detection is more than 98%,which can realize the rapid and accurate detection of parts.The detection and recognition technology can be applied to similar parts manufacturing industry to complete the classification and detection tasks,which has important research value and practical significance in industry.