基于综合信息融合神经网络的轴承故障诊断  

Bearing fault diagnosis based on integrated information fusion neural network

作　　者：裴红蕾[1] PEI Hong-lei(School of Electromechanical and Information Engineering,Wuxi Vocational Institute of Arts&Technology,Yixing 214200,China)

机构地区：[1]无锡工艺职业技术学院机电与信息工程学院,江苏宜兴214200

出　　处：《机电工程》2022年第4期495-500,共6页Journal of Mechanical & Electrical Engineering

基　　金：江苏省宜兴市科技计划资助项目(2019SF08);江苏省教育科学家“十四·五”规划项目(D/2021/03/44)。

摘　　要：因轴承的工作环境恶劣,导致其故障多发,在对轴承故障进行快速诊断和定位时存在困难,为此,提出了一种基于综合信息融合神经网络的轴承故障智能诊断方法。首先,介绍了前置神经网络的工作原理,推导了前置神经网络的链接权值系数训练方法,制定了前置神经网络的算法流程;并基于D-S证据论和Dempster组合规则,设计了后置神经网络的故障诊断方法;然后,提出了基于综合信息融合神经网络的轴承故障诊断方法,完成了两种神经网络的优势融合;最后,为了对基于综合信息融合神经网络的轴承故障诊断方法的有效性进行验证,采用了美国凯斯西储大学的轴承实验数据进行了试验。研究结果表明:无后置融合模块的故障识别率均值为90.45%,无前置融合模块的故障识别率均值为89.93%,综合信息融合神经网络的故障识别率均值为99.33%;该结果证明,基于综合信息融合神经网络的诊断方法具有较高的故障识别准确率和较强的鲁棒性,将综合信息融合神经网络应用于轴承故障诊断是有效的。Due to the poor working environment of the bearing,its faults frequently occur,and it was difficult to quickly diagnose and locate the bearing fault.For this reason,an intelligent diagnosis method for bearing faults based on comprehensive information fusion neural network was proposed.Firstly,the working principle of pre-neural network was introduced,the link weight coefficient training method of pre-neural network was deduced,and the algorithm flow of pre-neural network was formulated.Based on D-S evidence theory and Dempster combination rule,a fault diagnosis method of post neural network was designed.Then,a bearing fault diagnosis method based on comprehensive information fusion neural network was proposed,so that the advantages of the two neural networks were fused.Finally,the experimental verification was carried out based on the bearing experimental data of Case Western Reserve University.The research results show that the average fault recognition rate without post fusion module is 90.45%,the average fault recognition rate without pre-fusion module is 89.93%,and the average fault recognition rate of comprehensive information fusion neural network is 99.33%.The above data show that the integrated information fusion neural network has the highest fault recognition accuracy and strong robustness,which verifies the effectiveness and accuracy of comprehensive information fusion neural network in bearing fault diagnosis.

关 键 词：轴承故障 智能诊断 前置神经网络算法 综合信息融合 故障识别率 鲁棒性 

分 类 号：TH17[机械工程—机械制造及自动化] TH133.3