基于改进YOLOv8n的3D打印实时异常诊断算法  

Real-Time Anomaly Diagnosis Algorithm of 3D Printing Based on Improved YOLOv8n

作　　者：金凯 周敏[1,2,3] 胡佳乐 李欢 赵松怀[1,2,3] JIN Kai;ZHOU Min;HU Jiale;LI Huan;ZHAO Songhuai(Key Laboratory of Metallurgical Equipment and Control Technology Ministry of Education,Wuhan University of Science and Technology,Wuhan Hubei 430081,China;Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering,Wuhan University of Science and Technology,Wuhan Hubei 430081,China;Precision Manufacturing Institute,Wuhan University of Science and Technology,Wuhan Hubei 430081,China)

机构地区：[1]武汉科技大学,冶金装备及其控制省部共建教育部重点实验室,湖北武汉430081 [2]武汉科技大学,机械传动与制造工程湖北省重点实验室,湖北武汉430081 [3]武汉科技大学精密制造研究院,湖北武汉430081

出　　处：《机床与液压》2025年第7期177-183,共7页Machine Tool & Hydraulics

基　　金：国家自然科学基金面上项目(51975431)。

摘　　要：针对3D打印过程中异常诊断实时性不足和准确度不高的问题,提出一种改进的YOLOv8n模型(DSW-YOLOv8n)。在骨干网络中引入动态蛇形卷积(DSConv),增强网络对3D打印中出现的拉丝等细长弯曲局部结构特征的提取能力。增加小目标检测层并融入SA注意力机制,提升对小目标的异常检测能力。针对3D打印实时捕获图像质量低的问题,引入动态调整边界的Wise-IoU(WIoUv3)损失函数,降低对距离和纵横比等几何因素的惩罚,从而提高检测精度。最后,通过搭建实验平台,对所提模型进行性能验证。结果表明:DSW-YOLOv8n模型对3D打印异常检测精度和速度均优于Faster R-CNN、SSD和YOLOv5s等主流检测方法,其精度均值(mAP)达到了90.3%,较原始YOLOv8n模型提高了2.8%,平均帧率达到113帧/s,满足实时检测需求。In order to solve the problem of insufficient real-time and low accuracy of anomaly diagnosis in 3D printing process,an improved YOLOv8n model(DSW-YOLOv8n)was proposed.Dynamic serpentine convolutional(DSConv)was introduced into the backbone network to enhance the network′s ability to extract long and thin local structural features such as wire drawing in 3D printing.The small target detection layer was added and the SA attention mechanism was incorporated to improve the anomaly detection ability of small targets.To solve the problem of low quality of 3D printing images captured in real-time,the Wise-IoU(WIoUv3)loss function with dynamic boundary adjustment was introduced to reduce the penalty to geometric factors such as distance and aspect ratio,so as to improve the detection accuracy.Finally,the performance of the proposed model was verified by setting up an experimental platform.The results show that the proposed DSW-YOLOv8n model is superior to other mainstream detection methods such as Faster R-CNN,SSD and YOLOv5s for 3D printing anomaly detection accuracy and speed.The average accuracy(mAP)of the DSW-YOLOv8n model is 90.3%,which is 2.8%higher than that of the original YOLOv8n model,and the average frame rate reaches 113 frames/s,which meets the requirements of real-time detection.

关 键 词：3D打印 实时异常检测 YOLOv8n 动态蛇形卷积 小目标检测层 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP391[自动化与计算机技术—计算机应用技术]