极限学习机在航空轴承故障诊断中的应用研究  

Applied Research of Aero-Engine Bearing Fault Diagnosis Based on Extreme Learning Machine

作　　者：皮骏 马圣[2] 王力平 李章萍[4] PI Jun;MA Sheng;WANG Li-ping;LI Zhang-ping(College of General Aviation,Civil Aviation University of China,Tianjin 300300,China;College of Aeronautical Engineering,Civil Aviation University of China,Tianjin 300300,China;TE90,MTU Maintenance Zhuhai Co.Ltd,Guangdong Zhuhai 519030,China;Economics and Management College,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学通航学院,天津300300 [2]中国民航大学航空工程学院,天津300300 [3]珠海摩天宇航空发动机维修有限公司,TE90,广东珠海519030 [4]中国民航大学,经济与管理学院,天津300300

出　　处：《机械设计与制造》2020年第10期94-97,共4页Machinery Design & Manufacture

基　　金：中央高校基本科研业务费民航大学专项(3122013H001)。

摘　　要：基于实验平台,采集滚动轴承正常、内环故障、外环故障和滚珠故障四种工况的振动信号,利用时域分析法提取故障特征量。分析隐含层神经元数量、隐含层激活函数和样本比例对极限学习机网络诊断效果的影响,同时从网络对样本比例的适应性、算法的稳定性、仿真耗时和抗噪能力四方面比较BP、SVM和RBF网络。结果表明:针对轴承故障诊断,极限学习机在神经元数量较少时采用Sigmoid()函数、神经元数量较多时采用Hardlim()函数,其诊断效果较佳;极限学习机相比BP、SVM和RBF网络,能够更好的适应样本比例的变化,且算法的稳定性和准确性均为最优;极限学习机仿真计算时间相对较短、抗噪能力较强。Aero-engine bearing fault was diagnosis based on extreme learning machine(ELM)network in this paper.Based on the experimental platform,vibration signals of four conditions,including normal rolling bearing,inner ring fault,out ring fault and ball fault,were collected.And fault feature quantity of vibration signals was extracted by time domain analysis.ELM’s hidden layer neuron number,hidden layer activation function and sample proportion were analyze due to its effect network diagnosis.Then,the aspects of adaptability of network to sample proportion,stability of algorithm,computational times and anti-noise ability as the considerate factors to compare ELM,BP,SVM and RBF network.The results show that:based on bearing fault diagnosis,ELM adopts Sigmoid()function when the number of hidden neurons is less and Hardlim()function when there are more hidden neurons,its diagnosis effect is better.ELM can adapt to change of the bearing fault sample proportion better than BP,SVM and RBF network,and the stability and accuracy of the algorithm are all optimal.The ELM simulation computational times is relatively short and anti-noise ability is strong.

关 键 词：航空发动机 极限学习机 轴承 故障诊断 振动信号 时域分析 

分 类 号：TH16[机械工程—机械制造及自动化] TH17