摘要
超强耐热齿轮轴承表面因耐热材料特殊性质,使其表面形成随机分布的微小凸起、错综复杂的线条等独特纹理结构,而微小缺陷与复杂纹理特征差异较小,难以准确区分,易误判或掩盖真正缺陷,影响微小差异纹理捕捉及不同类型缺陷分类能力,导致缺陷检测准确性差。为此,提出改进CNN的超强耐热齿轮轴承表面缺陷检测方法。通过基于像素的马尔夫随机场算法获取像素特征及似然函数;利用随机区域合并算法提取轴承表面区域特征及似然函数;使用最大梯度算法捕获轴承表面图像边缘特征及似然函数;融合全部似然函数区分复杂纹理和真正缺陷特征,引入能量最小准则分割缺陷区域。最后,改进CNN分类器,利用AdaBoost构建AdaBoost-SVM级联分类器,识别各类缺陷。将分割后的缺陷区域输入训练后的级联强分类器,完成超强耐热齿轮轴承表面缺陷检测。实验结果表明,该方法可显著提升轴承表面缺陷检测结果的准确性。
Due to the special properties of heat-resistant materials,the surface of super heat-resistant gear bearings forms unique texture structures such as randomly distributed small protrusions and intricate lines.However,the differences between small defects and complex texture features are small,making it difficult to accurately distinguish and easily misjudge or conceal true defects,which affects the ability to capture small differences in texture and classify different types of defects,resulting in poor defect detection accuracy.Therefore,an improved CNN method for detecting surface defects in super heat-resistant gear bearings is proposed.Obtain pixel features and likelihood functions through the pixel based Markov random field algorithm.Extracting surface features and likelihood functions of bearings using random region merging algorithm.Capture edge features and likelihood function of bearing surface images using maximum gradient algorithm.Integrating all likelihood functions to distinguish complex textures from true defect features,and introducing the minimum energy criterion to segment defect areas.Finally,improve the CNN classifier and use AdaBoost to construct an AdaBoost SVM cascade classifier to identify various types of defects.Input the segmented defect areas into the trained cascaded strong classifier to complete the surface defect detection of the super heat-resistant gear bearing.The experimental results show that this method can significantly improve the accuracy of bearing surface defect detection results.
作者
苏靖
渠慎明
SU Jing;QU Shenming(College of Computer Engineering,Shangqiu Polytechnic,He’nan Shangqiu 476000,China;School of Software,He’nan University,He’nan Kaifeng 475004,China)
出处
《机械设计与制造》
北大核心
2026年第2期260-265,共6页
Machinery Design & Manufacture
基金
河南省高等学校重点科研项目计划支持(26B520040)
商丘市科技攻关项目(2024038)
河南省高等教育教学改革研究与实践项目(2024SJGLX0810)。
关键词
改进CNN
超强耐热齿轮
轴承表面
缺陷检测
Improve CNN
Super Heat-Resistant Gear
Bearing Surface
Defect Detection
