【作者】 赵鹏；

【导师】 吴晟； 赵栋梁；

【作者基本信息】 哈尔滨工程大学 ， 电子信息（专业学位）， 2025， 硕士

【摘要】 随着电子信息产业的迅速进步,特别是大规模及超大规模集成电路的广泛普及,电路对高频率稳定性的要求持续提升,石英晶振器件的应用范围也在不断拓展。作为电路系统里基本频率信号的核心供给部件,石英晶振器件一旦运转异常,会对整个设备及系统产生显著的不良影响。在生产过程中,这类器件的表面可能出现表面破损、面胶缺失或涂抹不均、划痕等多种瑕疵,直接影响其使用寿命与工作效能。面对当前大规模集成化和自动化的生产模式,传统依靠人工肉眼观察的检测方式,已难以满足高效的检测标准。有鉴于此,本文构建了一套基于机器视觉技术的石英晶振器件表面缺陷检测系统,致力于研发配套的缺陷检测算法,以此取代现有的人工检测手段,实现快速、精确且稳定的表面缺陷检测,进而提高产品合格比例并推动批量化生产的自动化进程。本文针对基于机器视觉的石英晶振器件表面缺陷检测系统,提出了涵盖机械设备端、上位机端和电气供应端的整体设计方案,并明确了系统的运行流程。在硬件构成方面,系统包含自动上下料装置、基板传输装置、图像采集装置以及缺陷识别装置。其中,图像采集模块选用高分辨率工业相机、工业镜头以及适配光源,通过与PLC、单片机和工控机相结合,实现设备控制及上位机的部署工作。针对传统检测方法存在检测时间长和检测精度低等问题,本文提出了一种基于改进YOLOv8s模型的缺陷检测算法。改进后的YOLOv8s模型中引入了卷积块注意力模块(Convolutional Block Attention Module,CBAM)机制,该模块通过提升模型的特征提取和特征融合能力,加强了对小目标缺陷的检测。此外,添加了专门的小目标检测层,有效解决了因石英晶振器件缺陷数据集中小目标比例较高而产生的漏检问题。同时为了降低由于引入CBAM机制和小目标检测层所增加的计算量与训练时长,采用改进后的轻量化网络Mobile Net V3替代了YOLOv8s原始主干网络。改进后模型的参数量降至7.3M,计算量显著减少,同时检测精度提升,平均精度均值达到0.970,召回率为0.958。在石英晶振器件的大规模集成生产环境下,本文改进后的模型在检测精度、稳定性和速度上表现出良好的性能。该检测系统能高效完成石英晶振器件表面缺陷的检测任务,可以显著提升生产效率和产品质量,可有效替代人工检测,适用于自动化生产。更多还原

【Abstract】 With the rapid development of the electronic information industry,especially the widespread adoption of large-scale and very-large-scale integrated circuits,the demand for high-frequency stability in current circuits is continuously increasing.As a result,the application scope of quartz crystal oscillator devices has been expanding.These devices serve as the key components for supplying fundamental frequency signals in circuit systems.Any abnormal operation of quartz crystal oscillator devices can have significant negative impacts on the entire equipment and system.During the manufacturing process,various defects may occur on the surface of these devices,such as surface damage,missing or uneven surface adhesive,and scratches,all of which directly affect their service life and operational efficiency.In the context of current large-scale integrated and automated production models,the traditional manual visual inspection method can no longer meet the requirements of efficient and accurate detection.Therefore,this paper constructs a surface defect detection system for quartz crystal oscillators based on machine vision technology and focuses on developing corresponding defect detection algorithms.The aim is to replace the existing manual inspection method,achieve fast,accurate,and stable surface defect detection,increase the product pass rate,and promote the mass production process.This paper proposes an overall design scheme covering the mechanical equipment end,upper computer end,and electrical supply end for the surface defect detection system of quartz crystal oscillators based on machine vision,and clarifies the system’s operation process.In terms of hardware composition,the system includes an automatic loading and unloading device,a substrate transmission device,an image acquisition device,and a defect recognition device.The image acquisition module selects high-resolution industrial cameras,industrial lenses,and suitable light sources.By combining with PLC,single-chip microcomputers,and industrial computers,it realizes the control of equipment and the deployment of the upper computer.Aiming at the problems of long detection time,low detection accuracy,and unclear defect features in traditional detection methods,this paper proposes a defect detection algorithm based on an improved YOLOv8s model,and the detection results are fed back to the production line in real time.The improved YOLOv8s model incorporates the CBAM attention mechanism.This module enhances the model’s ability to extract and fuse features,thereby strengthening the detection of small-target defects.In addition,a special small-target detection layer is added,effectively addressing the problem of missed detection caused by the high proportion of small targets in the datasets.To reduce the increased computational load and training time due to the introduction of the CBAM mechanism and the small-target detection layer,the improved lightweight network Mobile Net V3 replaces the original backbone network of YOLOv8s.After the improvement,the model’s parameter quantity is reduced to 7.3M,and the computational load is significantly decreased.Meanwhile,the detection accuracy is improved,with m AP@0.5 reaching 0.970,and the recall rate is 0.958.Experimental results show that in a large-scale integrated production environment,the improved model performs well in terms of detection accuracy,stability,and speed.This detection system can efficiently complete the task of detecting surface defects of quartz crystal oscillators,significantly improving production efficiency and product quality,making it suitable for industrial applications.更多还原