金属嵌入件装配缺陷视觉检测技术研究  

Research on Vision Detection Technology for Defects in Display Backplate Embedded Component Assembly

作　　者：高岩 方成刚[1] GAO Yan;FANG Chenggang(School of Mechanical and Power Engineering,Nanjing Tech University,Nanjing 211800,China)

机构地区：[1]南京工业大学机械与动力工程学院,南京211800

出　　处：《组合机床与自动化加工技术》2024年第12期169-174,共6页Modular Machine Tool & Automatic Manufacturing Technique

摘　　要：针对目前显示器背板上金属嵌入件装配缺陷依赖人工目视检测、效率低、主观性大的问题,提出一种显示器背板金属嵌入件装配缺陷视觉检测方法。通过预处理方法得到高质量的装配区域图像,采用蜣螂优化算法(dung beetle optimizer,DBO)优化二维Otsu,分割出嵌入件,提取嵌入件的形状、位置特征并对特征进行归一化融合,输入到DBO-BP神经网络中,对装配缺陷进行识别。实验结果表明,该方法在数据集上准确率可达97.5%,在分割和分类问题上采用DBO优化的二维Otsu和BP神经网络表现均优于采用粒子群算法(particle swarm optimization,PSO)和遗传算法(genetic algorithm,GA)的优化。所提出的方法能够有效检测出装配后的缺陷,对工厂的噪声与光照条件变化有一定适应性,满足企业生产的实时要求。To address the issue of low efficiency and high subjectivity in manual visual inspection of metal insert assembly defects on display backplates,this paper proposes a visual inspection method for detecting assembly defects of metal inserts on display backplates.By using preprocessing techniques,high-quality images of the assembly area are obtained,and the dung beetle optimizer(DBO)is applied to optimize the two-dimensional Otsu method to segment the inserts.The shape and position features of the inserts are extracted and normalized for fusion,and then input into the DBO-optimized BP neural network for defect recognition.Experimental results show that this method achieves an accuracy of 97.5%on the dataset,and the performance of using DBO-optimized two-dimensional Otsu and BP neural network is superior to the optimization using particle swarm optimization(PSO)and genetic algorithm(GA)for both segmentation and classification problems.The proposed method can effectively detect assembly defects and is adaptable to factory noise and lighting condition changes,meeting the real-time requirements of enterprise production.

关 键 词：装配缺陷 二维OTSU 特征提取 DBO-BP神经网络 

分 类 号：TH165[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]