样本不均衡情况下的航空发动机轴承故障诊断方法  

Research on Fault Diagnosis Strategy for Aeroengine Bearing with Imbalanced Data

作　　者：王海泉 王亚辉 杨岳毅 Kurkova Olga Petrovna 温盛军 陈乐瑞 路成钢 WANG Haiquan;WANG Yahui;YANG Yueyi;Kurkova Olga Petrovna;WEN Shengjun;CHEN Lerui;LU Chenggang(Zhongyuan-Petersburg Aviation College/College of Intelligent Sensing and Instrument,Zhongyuan University of Technology,Zhengzhou 450007,China;School of Automation and Electrical Engineering,Zhongyuan University of Technology,Zhengzhou 450007,China;School of Electronic and Laser Instrument,Saint Petersburg State University of Aerospace Instrumentation,Saint Petersburg,14-51)

机构地区：[1]中原工学院中原彼得堡航空学院(智能感知与仪器学院),河南郑州450007 [2]中原工学院自动化与电气工程学院,河南郑州450007 [3]俄罗斯圣彼得堡国立宇航与仪器制造大学电子与激光仪器系,圣彼得堡14-51

出　　处：《郑州航空工业管理学院学报》2024年第4期5-11,共7页Journal of Zhengzhou University of Aeronautics

基　　金：河南省高校青年骨干教师项目(G2021026006L);河南省高端外籍专家项目(HNGD2024032);河南省高校重点科研项目(24A413013)

摘　　要：航空发动机滚动轴承工作在高温、高压、高转速的恶劣环境下,故障率高,因此对于其故障的准确识别和判断尤为重要。但由于轴承故障的偶发性,各类故障样本不均衡的问题非常突出,大大影响了基于数据的模式识别方法的准确性。本文提出了一种样本不均衡条件下的航空发动机滚动轴承智能诊断方法,采用合成少数类过采样方法进行样本平衡,在完成时频域特征提取和特征选择之后,利用改进蜂群算法优化后的随机森林策略实现轴承故障分类,并在凯斯西储大学轴承数据集和实验室构建的模拟航空发动机滚动轴承数据集上进行了实验验证。结果显示在凯斯西储大学中度不均衡数据集下,故障识别的准确率为98.30%,在重度不均衡数据集下的分类结果为96.30%。在实验室构建的模拟航空发动机滚动轴承实验台上,不均衡数据集的分类结果为97.65%,重度不均衡数据集的分类准确率为95.67%。相关实验证明本文所提算法能够有效完成不均衡样本下的航空发动机滚动轴承故障诊断任务。The mechanical rolling bearing of aeroengine always works in poor environment with long running time and high load,which causes high failure rate and threatens the safety of people’s live,so it is important to early detect the faults of rolling bearings.But as the faults occurred randomly,the fault data is unbalanced,and the accuracy of fault detection with pattern recognition strategy will be affected.To solve this problem,an intelligent diagnosis method for aeroengine rolling bearing with unbalanced samples is proposed in this paper.Firstly,the reduced-order variational mode decomposition is used to balance the fault data,and random forest algorithm is introduced to recognize the faults,and in order to improve the detection accuracy,an improved artificial bee colony algorithm is proposed to optimize the random forest strategy.Finally,the proposed strategy is verified on the Case Western Reserve University bearing dataset and the platform constructed by our laboratory.The accuracy is 98.30%in the moderately imbalanced dataset of Case Western Reserve University,and the classification result is 96.30%in the severely imbalanced dataset.With the platform in our laboratory,the classification result for moderate imbalanced data set is 97.65%,and the classification accuracy for severely imbalanced data set is 95.67%.

关 键 词：航空发动机轴承 故障诊断 数据不平衡 过采样 蜂群算法 

分 类 号：TM931[电气工程—电力电子与电力传动]