废旧零/部件批量表面失效形式识别与分类方法研究  

Identification and classification method of surface failure forms for mass retired parts

作　　者：夏绪辉[1] 周诚 王蕾[1] 张泽琳 刘翔[1] XIA Xuhui;ZHOU Cheng;WANG Lei;ZHANG Zelin;LIU Xiang(Key Laboratory of Metallurgical Equipment and Control Technology,Wuhan University of Science and Technology,Ministry of Education,Wuhan 430081,China)

机构地区：[1]武汉科技大学冶金装备及控制教育部重点实验室,武汉430081

出　　处：《现代制造工程》2021年第8期147-154,共8页Modern Manufacturing Engineering

基　　金：国家自然科学基金项目(51805383);湖北省重点研发计划项目(2020BAB047);襄阳市重点研发计划项目(2020AAT001420)。

摘　　要：按失效形式对大规模废旧零/部件进行预分类,是提高废旧零/部件批量再制造效率与效益的重要保障。针对大批量废旧零/部件表面失效人工识别效率低、漏检率和错检率高,导致难以满足自动化在线检测与分类需求的问题,提出一种基于机器视觉的废旧零/部件批量在线表面失效形式识别与分类方法。在分析再制造检测服务概念与废旧零/部件失效形式的基础上,针对图像视觉下废旧零/部件“近形-异类”表面失效形式误判率高的问题,利用ROI高斯学习策略对废旧零/部件表面失效区域精准定位,提取候选分类特征,利用遗传算法(Genetic Algorithm,GA)筛选出关键分类特征,采用支持向量机(Library for Support Vector Machines,LIBSVM)建立失效形式分类模型,并通过K折交叉验证方法(K-fold Cross-Validation,K-CV)对其惩罚因子和核参数进行优化。以某退役齿轮零件为例对该方法的有效性与可行性进行验证,结果显示:该方法对再制造回收零/部件失效形式的分类精度达到96.7%,比同类算法精度提高了2.3%,比熟练人工检测精度提高了2.5%,表明该方法不仅具有一定的理论优越性,而且具有广阔的应用前景。It is an important guarantee to improve the efficiency and benefit of batch remanufacturing of retired parts by pre-classification of large scale retired parts according to failure forms.Aiming at the problems of low efficiency of manual identification of surface failure of large quantities of retired parts and high failure rate of omission and error,which make it difficult to meet the requirements of automatic online detection and classification,a method of online surface failure identification and classification of large quantities of retired parts based on machine vision was proposed.Based on the analysis of the concept of remanufacturing testing service and the failure types of retired parts,it is easy for the surface failure analysis of retired parts under image vision to have a high misjudgment rate of“similar shape and different type”.By using ROI Gaussian learning strategy to accurately position the retired parts surface failure area,candidate classification features were extracted.By using the Genetic Algorithm(GA),the key classification feature were selected.By using Library for Support Vector Machine(LIBSVM),a failure pattern classification model was established,and by the method of K-fold Cross-Validation(K-CV),the punishment factor and kernel parameter was optimized.The effectiveness and feasibility of this method were verified by taking a retired gear part as an example.The results show that the classification accuracy of the method was 96.7%,2.3%higher than that of the similar algorithm,2.5%higher than that of the skilled manual detection,indicating that the method has not only certain theoretical superiority,but also broad application prospect.

关 键 词：再制造 失效检测 分类模型 支持向量机 

分 类 号：TH161[机械工程—机械制造及自动化]