基于深度学习的电机外观缺陷智能检测系统设计  

Design of intelligent detection system for motor appearance defects based on deep learning

作　　者：刘志昌 LIU Zhichang(Guangdong Provincial Key Laboratory of High Performance Servo System,Zhuhai 519000,China;GREE Electric Appliances Inc.of Zhuhai,Zhuhai 519000,China)

机构地区：[1]广东省高性能伺服系统企业重点实验室,珠海519000 [2]珠海格力电器股份有限公司,珠海519000

出　　处：《自动化与仪器仪表》2023年第5期314-317,共4页Automation & Instrumentation

基　　金：广东省高性能伺服系统企业重点实验室(2020B121202017)。

摘　　要：针对电机外观缺陷检测通过人工检查方法准确率偏低、误判率高、严重影响电机出厂质量等问题,通过分析电机外观缺陷类型,设计了一种基于深度学习方法的智能视觉检测系统,该系统方案由训练服务器、工业视觉控制器、工业相机、LED光源和控制器以及数字IO模块构成,融合了深度学习与传统CV视觉检测算法,采用了全新的系统架构与快速部署设计方案,实现了自动化、高精度的视觉识别检测,在效率比人工高几倍的同时,还能通过增量训练不断提升推理的准确率。尤为重要的是,实现这一效果仅需要在通用CPU平台上实现,进一步降低了成本,提升产线运行效率和质量。In view of the problems of low accuracy,high false positive rate and serious impact on the quality of the motor factory through manual inspection method,through the analysis of the type of motor appearance defects,an intelligent visual inspection system based on deep learning method is designed,which is composed of training server,industrial vision controller,industrial camera,LED light source and controller,and digital IO module,integrating deep learning and traditional CV visual inspection algorithm,adopting a new system architecture and rapid deployment design scheme.It realizes automated,high-precision visual recognition detection,and while the efficiency is several times higher than that of humans,the accuracy of inference can be continuously improved through incremental training.In particular,this effect only needs to be achieved on a general-purpose CPU platform,further reducing costs and improving the efficiency and quality of production line operations.

关 键 词：缺陷检测 工业视觉 自动化 传统视觉检测 图像分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]