基于自适应深度卷积神经网络的发射车滚动轴承故障诊断研究  

Rolling bearing fault diagnosis of launch vehicle based on adaptive deep CNN

作　　者：曹继平 王赛 岳小丹[2] 雷宁 CAO Jiping;WANG Sai;YUE Xiaodan;LEI Ning(School of Combat Support,Rocket Force University of Engineering,Xi’an 710025,China;Troop 96884,PLA,Luoyang 471000,China)

机构地区：[1]火箭军工程大学作战保障学院,西安710025 [2]中国人民解放军96884部队,河南洛阳471000

出　　处：《振动与冲击》2020年第5期97-104,149,共9页Journal of Vibration and Shock

摘　　要：作为发射车的关键组成部件,滚动轴承的工作环境复杂,故障诊断困难。提出一种自适应深度卷积神经网络,针对传统CNN诊断方法存在的计算效率较低、参数调试需人工经验指导等问题,采用粒子群优化算法确定CNN模型结构和参数,应用主成分分析法将故障诊断特征学习过程可视化,评估其特征学习能力。将提出方法应用于发射车滚动轴承故障诊断,对比标准CNN、SVM、ANN诊断方法,10种工况的诊断结果表明,提出方法诊断精度高且鲁棒性好。As key components of launching vehicle,rolling bearings’working conditions usually are very complex to make their fault diagnosis be difficult.Here,in order to effectively perform rolling bearing fault diagnosis,a novel method called the adaptive deep convolutional neural network(CNN)was proposed.Aiming at problems of lower calculation efficiency and parametric adjusting needing manual experience existing in the traditional CNN diagnosis method,PSO algorithm was used to determine structure and parameters of a CNN model.The principal component analysis(PCA)method was used to visualize its fault diagnosis feature learning process,and evaluate its feature learning ability.The diagnosis results with several diagnosis methods,respectively under 10 different bearing working conditions showed that compared with standard CNN,SVM and ANN diagnosis methods,the proposed method has higher diagnosis accuracy and better robustness.

关 键 词：故障诊断 卷积神经网络 粒子群优化 发射车 滚动轴承 特征学习 

分 类 号：TP206.3[自动化与计算机技术—检测技术与自动化装置]