改进YOLOv8的汽车制动器装配缺陷检测算法  

Automotive Brake Assembly Defect Detection on Improved YOLOv8

作　　者：叶娜 周学良 张映锋 梁宏 YE Na;ZHOU Xueliang;ZHANG Yingfeng;LIANG Hong(School of Mechanical Engineering,Hubei University of Automotive Technology,Shiyan 442002,China;Hubei Huayang Vehicle Braking Co.,Ltd.,Shiyan 442012,China)

机构地区：[1]湖北汽车工业学院机械工程学院,十堰442002 [2]湖北华阳汽车制动器股份有限公司,十堰442012

出　　处：《组合机床与自动化加工技术》2025年第4期140-145,151,共7页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：省部共建精密电子制造技术与装备国家重点实验室开放课题项目(JMDZ202321)。

摘　　要：针对汽车制动器在装配过程中易出现各类小零件漏装、错装和装配不规范等问题,提出了一种基于YOLOv8网络改进的机器视觉检测算法(Brake-YOLO),用于汽车制动器装配缺陷检测。首先,提出了高效多尺度卷积注意力模块C2f_EMA对视觉图像全局信息进行编码,以提高模型的特征提取能力与效率,以及对小目标的检测能力;其次,引入自适应空间特征融合结构(ASFF)以减少特征融合过程中的信息丢失,增强模型特征融合能力;最后,采用均衡焦点损失函数(EFL Loss)改善样本类别不平衡问题,以提高对汽车制动器装配缺陷的检测准确性和鲁棒性。实验结果表明,改进后的算法相比原算法在精确度、召回率、均值平均精度上分别提高了2.8%、1.2%、3.5%,检测速度达到了145.8 FPS(每秒帧数),进一步提高了汽车制动器装配缺陷检测的精度和效率。Aiming at the automobile brake in the assembly process is prone to various types of small parts leakage,misassembly and assembly irregularities,etc.,a machine vision inspection algorithm based on the YOLOv8 network improvement is proposed(Brake-YOLO).First,the adaptive convolutional attention module C2f_EMA is proposed to encode the global information of visual images to improve the feature extraction capability and efficiency of the model,as well as the detection of small targets;Second,the Neck architecture is optimized by combining the reparameterized generalized feature pyramid network(RepGFPN)structure to extract contextual information more efficiently in order to enhance the model′s multi-scale feature fusion capability;Finally,equalized focus loss function(EFL Loss)is used to improve the sample category imbalance problem in order to improve the accuracy and robustness of detecting defects in automotive brake assembly.The experimental results show that the improved algorithm improves 2.8%,1.2%,and 3.5%in precision,recall,and mean average precision,respectively,compared with the original algorithm,and the detection speed reaches 145.8 FPS(frames per second),which further improves the precision and efficiency of the detection of defects in the automotive brake assembly.

关 键 词：汽车制动器 装配缺陷检测 深度学习 YOLO 多尺度特征融合 

分 类 号：TH165[机械工程—机械制造及自动化] TG66[金属学及工艺—金属切削加工及机床]