滚动轴承变工况状态评估的特征融合技术研究

发布时间：2018-05-23 17:37

本文选题：航空发动机滚动轴承 + 故障诊断　；参考：《南京航空航天大学》2016年硕士论文

【摘要】：滚动轴承作为航空发动机关键部件直接影响着飞行安全,对滚动轴承进行状态检测,尽早发现轴承的故障征兆,对于有效减少飞行事故的发生,实施滚动轴承剩余寿命预测具有重要意义。现有的滚动轴承状态评估较少考虑载荷和转速变化对轴承状态评估的影响,缺乏对特征灵敏度的研究和融合,对早期故障检测不灵敏,对轴承状态评估不准确,因此需要进行变工况下多特征融合。本文研究了航空发动机滚动轴承在变工况状态下的多特征提取与融合技术。主要研究工作体现在:(1)提出了三种多特征融合方法,即距离判别法、一类分类法和后验概率法。距离判别法是利用滚动轴承运行的正常数据的振动特征进行欧氏距离学习,并对未知状态与正常状态的距离进行比较。一类分类法是对正常数据样本的分布做出正确的描述,检验对未知样本的分类就是检验其是否服从分布。后验概率法是基于后验概率的支持向量机算法,使用正常状态和严重故障状态的样本数据,形成训练样本,对后验概率支持向量机进行学习,既可以实现分类问题,又可以结合贝叶斯决策规则实现分类结果的概率估计。(2)进行滚动轴承单点故障模拟试验,得到4组振动加速度信号,从时域、频域和时频域中提取出了13个无量纲特征,进行了13个特征的灵敏度分析。提取转速信号,比较了不同转速对故障特征灵敏度的影响。利用特征融合方法,对多维特征进行了融合,试验结果充分验证了方法的正确性。(3)进行滚动轴承性能退化试验,得到航空滚动轴承在整个从正常到异常状态下的不同工作状态的振动试验数据。利用本文特征融合方法进行了特征融合和状态评估,结果证明:提取的多维特征通过融合后进行状态评估能明显区分出轴承正常与异常状态,即本文的特征融合状态评估方法有很好的工程应用价值。
[Abstract]:As the key component of aero-engine, rolling bearing has a direct impact on flight safety. To detect the status of rolling bearing and find the bearing fault as soon as possible can effectively reduce the occurrence of flight accident. It is of great significance to predict the remaining life of rolling bearing. The present status evaluation of rolling bearings seldom considers the influence of load and rotational speed change on bearing state evaluation, and lacks the research and fusion of characteristic sensitivity, which is insensitive to early fault detection and inaccurate for bearing state evaluation. Therefore, it is necessary to perform multi-feature fusion under variable operating conditions. In this paper, the multi-feature extraction and fusion technology of aeroengine rolling bearing under variable working condition is studied. In this paper, three multi-feature fusion methods are proposed, which are distance discrimination, classification and posteriori probability. The distance discriminant method is based on the vibration characteristics of the normal data of rolling bearing operation to study the Euclidean distance and to compare the distance between unknown state and normal state. A kind of classification is to describe the distribution of normal data samples correctly, and to test the classification of unknown samples is to test whether they are subordinate to the distribution. A posteriori probability method is a support vector machine algorithm based on a posteriori probability, which forms a training sample by using the sample data of the normal state and the serious fault state, and studies the posteriori probabilistic support vector machine, which can realize the classification problem. In addition, the probability estimation of classification results based on Bayesian decision rule can be used to simulate the single point fault of rolling bearing. Four sets of vibration acceleration signals are obtained, and 13 dimensionless features are extracted from time domain, frequency domain and time frequency domain. Sensitivity analysis of 13 features was carried out. The effect of different rotational speed on fault characteristic sensitivity was compared by extracting rotational speed signal. The feature fusion method is used to fuse the multi-dimensional features. The experimental results fully verify the correctness of the method. (3) the rolling bearing performance degradation test is carried out. Vibration test data of aeronautical rolling bearings in different working states from normal to abnormal state are obtained. The feature fusion and state evaluation are carried out by using the method of feature fusion in this paper. The results show that the extracted multidimensional features can distinguish the normal and abnormal states of bearings obviously by the state evaluation after fusion. That is to say, the method of feature fusion state evaluation in this paper has good engineering application value.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：V263.6

【参考文献】