基于深度学习的沉孔垫圈装配情况检测方法  

Detection Method for Assembly Status of Countersunk Gasket Based on Deep Learning

作　　者：刘宇旭云 孙宏博 董浩 杨东升 LIU Yuxuyun;SUN Hongbo;DONG Hao;YANG Dongsheng(Guobo Electronics Co.,Ltd.,Nanjing Jiangsu 210011)

机构地区：[1]南京国博电子股份有限公司,江苏南京210011

出　　处：《软件》2024年第9期97-99,共3页Software

摘　　要：针对沉孔螺栓自动装配过程中孔内垫片的有无检测问题,本文提出了一种基于深度学习的视觉检测方法,利用改进的YOLOv8算法模型对不同表面性质的产品进行沉孔内垫片检测。首先,对不同产品沉孔中垫片构建了垫片装配情况数据集;其次,使用GhostNet网络改良原有的YOLOv8主干网络,并通过新增第六个深度学习计算层,在不影响检测效果的同时使模型得到轻量化;最后,采用SmoothL1损失函数,提高了模型训练收敛速度。实验结果表明,优化后的模型精确度为99.5%、召回率为97.4%、Map@0.5为98.4%、推理耗时为6.4ms,相比YOLOv8原版本精确度提升了1.9%、召回率提升了11.8%、Map@0.5提升了13%。模型投入实际生产后,垫片装配不良漏检率0%、错杀率0.25%,对解决螺栓自动装配场景下垫片装配异常检测问题具有重要参考价值。This paper proposes a visual inspection method based on deep learning to address the issue of detecting the presence or absence of gaskets inside the holes during the automatic assembly process of countersunk bolts.The improved YOLOv8 algorithm model is used to detect gaskets inside the holes of products with different surface properties.Firstly,a dataset of gasket assembly was constructed for gaskets in different product countersunk holes;Secondly,the GhostNet network is used to improve the original YOLOv8 backbone network,and a sixth deep learning computing layer is added to make the model lightweight without affecting the detection performance;Finally,the SmoothL1 loss function was adopted to improve the convergence speed of the model training.The experimental results show that the optimized model has an accuracy of 99.5%and a recall rate of 97.4%Map@0.5 Compared to the original version of YOLOv8,YOLOv8 has improved accuracy by 1.9%and recall rate by 11.8%,with a reasoning time of 6.4ms and a rate of 98.4%Map@0.5 Increase by 13%.After the model was put into actual production,the defect detection rate of gasket assembly was 0%,and the error rate was 0.25%.This has important reference value for solving the problem of gasket assembly anomaly detection in the scenario of automatic bolt assembly.

关 键 词：深度学习 目标检测 垫圈检测 YOLOv8 损失函数 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]