基于BP神经网络与CNN算法结合的汽车轴承润滑油磨粒检测研究  

Study on Grinding Particle Detection of Automotive Bearing Lubricating Oil Based on BP Neural Network and CNN Algorithm

作　　者：王飞 何磊 闻有成 郑蕊 WANG Fei;HE Lei;WEN Youcheng;ZHENG Rui(College of Automotive Engineering,Anhui Vocational and Technical College,Hefei 230011,China;School of Engineering Science,University of Science and Technology of China,Hefei 230021,China;School of Economics and Technology,Anhui Agricultural University,Hefei 230036,China)

机构地区：[1]安徽职业技术学院汽车工程学院,合肥230011 [2]中国科学技术大学工程科学学院,合肥230021 [3]安徽农业大学经济技术学院,合肥230036

出　　处：《长春工程学院学报(自然科学版)》2025年第1期66-74,共9页Journal of Changchun Institute of Technology:Natural Sciences Edition

基　　金：安徽省高校科学研究重点项目(2022AH052068,2022AH052057)。

摘　　要：汽车轴承油液磨粒智能检测技术是汽车智能化综合检测发展的方向之一,针对汽车轴承智能磨粒识别存在的磨损粒子分类精度低、泛化能力差问题,提出了一种基于二分类有效识别磨粒的方法。用特定的粒子形态进行反向传播计算并运用到一级分类检测中,确定摩擦、切割、球形3种类型的磨粒,在二级分类中用CNN深度学习模型来区分疲劳和严重滑动磨粒。试验表明:BP神经网络与CNN算法结合对磨粒的识别率≥99%,改进的LeNet-5模型对磨粒的识别率为96%,相比BP神经网络的85%有明显提高。“预处理+反向传播算法与卷积神经网络结合”的图像识别方法为实现油液中磨粒的精准识别提供了理论支持。Intelligent detection technology for automotive bearing oil abrasive particles is one of the directions for the development of intelligent comprehensive detection of automobiles.In response of the problems of low classification accuracy and poor generalization ability of wear particles in intelligent abrasive particle recognition of automotive bearings,a method based on binary classification for effective identification of abrasive particles is proposed.It uses specific particle shapes for backpropagation calculations and applies them to primary classification detection to determine three types of abrasive paticles,friction,cutting,and spherical.In secondary classification,a CNN deep learning model is used to distinguish between fatigue and severe sliding abrasive particles.The experiment shows that the combination of BP neural network and CNN algorithm has a recognition rate of more than 99%for abrasive particles,and the improved LeNet-5 model has a recognition rate of 96%for abrasive particles,which is significantly improved compared with the previous 85%of the BP neural network.The image recognition method combining preprocessing backpropagation algorithm and convolutional neural network provides theoretical support for achieving accurate recognition of abrasive particles in oil.

关 键 词：BP神经网络 卷积神经网络 LeNet-5 汽车轴承 深度学习 

分 类 号：TH117[机械工程—机械设计及理论]