基于卷积神经网络的铆钉缺陷检测方法研究  

Diagnosis Method of Riveting Defects Based on Convolutional Neural Network

作　　者：邹恒毅 范周童 聂广银 蒲程 刘凉[1,2] Zou Hengyi;Fan Zhoutong;Nie Guangyin;Pu Cheng;Liu Liang

机构地区：[1]天津理工大学天津市先进机电系统设计与智能控制重点实验室,天津市300384 [2]机电工程国家级实验教学示范中心(天津理工大学),天津市300384

出　　处：《时代汽车》2024年第2期4-6,共3页Auto Time

基　　金：天津理工大学大学生创新创业训练计划项目(202110060100)。

摘　　要：针对铆接过程中铆钉内部断裂缺陷难以分类判别问题,利用卷积神经网络和长短期记忆网络设计检测算法。首先,使用平滑滤波法将特征信号中的异常噪声值滤去,同时根据拟合函数对检测信号进行插值,提高特征数据的信息量。其次,利用处理后的故障样本训练网络来实现铆钉缺陷诊断。实验结果表明:该方法总体识别准确率可达99%,能够有效地进行铆钉缺陷诊断。In view of the difficulty in classification and identification of internal fracture defects in rivets,a method was proposed based on convolutional neural network and Bi-LSTM.Firstly,the abnormal noise signal isfiltered by the smoothfiltering method.The detection signal is interpolated according to the appropriatefitting function to improve the amount of the feature data.Secondly,the detecting samples are treated as the inputs of the designed model.The experimental results show that the overall identification accuracy of the presented method is up to 99%,which can effectively diagnose the defects in rivets.

关 键 词：铆接质量 涡流检测 拟合插值 缺陷识别 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]