基于机器视觉的O型密封圈缺陷检测研究  

Defect detection of O-shape sealing ring based on machine vision

作　　者：杨怀斌 杨智才 刘胜祥 崔敏[2] YANG Huaibin;YANG Zhicai;LIU Shengxiang;CUI Min

机构地区：[1]罗定职业技术学院,广东罗定527200 [2]五邑大学,广东江门529000

出　　处：《现代机械》2025年第2期94-98,共5页Modern Machinery

基　　金：2022年度广东省普通高校青年创新人才类项目(2022KQNCX277);2021年广东省科技创新战略专项项目(“大专项+任务清单”)(2021020401);2022年广东省科技创新战略专项资金项目(pdjh202261083);2023年度云浮市科技创新项目(S2023040407)。

摘　　要：针对O型黑色密封圈人工检测效率低和成本高等问题,提出了一种采用机器学习的方法对黑色O型密封圈表面缺陷进行检测。一个正品密封圈表面是没有任何纹理特征即表面光滑,而有裂缝、缺胶、划痕等的密封圈是有纹路的,因此,针对这一点选用纹理特征作为特征向量,采用MLP神经网络作为分类器进行训练识别。通过实验证明此方法的检测平均准确率高达93.6%。此方法对缺胶、裂缝、划痕检测很实用,给密封圈的质量检测提供了一种可靠的方法。Aiming at the issues of low efficiency and high cost in manual inspection of the black O-shaped sealing ring,this article proposes a machine learning method to detect the surface defects of the black O-shaped sealing ring.For a genuine sealing ring,the surface has no texture features and is smooth,while the sealing ring with cracks,rubber starve and scratches has patterns.Therefore,this article selects texture features as the feature vectors,and uses MLP neural network as the classifier for training and recognition.Experimental results have shown that the average detection accuracy of this method is as high as 97.6%.This method is very practical for detecting rubber starve,cracks and scratches,providing a reliable method for quality inspection of sealing rings.

关 键 词：密封圈 机器学习 纹理特征 MLP神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]