基于神经网络和流形学习的军用航空发动机故障诊断技术研究

发布时间：2018-04-11 02:31

本文选题：军用航空发动机 + 滚动轴承　；参考：《吉林大学》2017年硕士论文

【摘要】：随着科学技术的发展,对于军用航空发动机的可靠性要求日益增高。然而由于功能的不断扩展,服役条件的苛刻,使得军用航空发动机机械零件的结构日趋复杂,发生故障的概率也越来越高。造成军用航空发动机零部件故障或失效的原因错综复杂,受到人为失误、材料缺陷、使用环境变化,以及疲劳、磨损和老化效应等很多因素的影响。这样就使军用航空发动机的故障诊断和分析越来越困难,需要更多的航空维修保障工作也要消耗大量人力物力去保障飞机的可靠性。传统的故障诊断技术虽然能够进行故障诊断,但是诊断的效率和准确率都不高。本文针对军用航空发动机故障诊断领域的迫切需求,应用数据挖掘的理论方法和技术,将神经网络和流形学习算法相结合应用到军用航空发动机的故障诊断。本文介绍了故障诊断技术和数据挖掘技术的国内外研究现状,以及军用航空发动机滚动轴承故障诊断所需的背景知识,包括基本的振动机理,常见的故障类型和传统的诊断方法,然后详细说明了滚动轴承振动信号的特征。本文详细讲解并比较神经网络在故障诊断领域常用的两种分类器PNN和SOM,通过实验分析验证了SOM分类器在故障诊断应用中效果更好。深入研究流形学习的理论和方法,分析了常见的三种流形学习算法:ISOMAP、LLE和LE。比较了传统的线性降维方法PCA和非线性降维方法LE,并提出LE算法的改进方案,用改进后的距离函数代替原来的欧式距离函数,能更好的反应高维数据集的局部结构信息,提高LE算法的性能。本文主要研究的是军用航空发动机滚动轴承的故障诊断问题,滚动轴承是军用航空发动机中最核心的零部件之一,在其中起着承受及传递载荷的作用,其运行状态对发动机的工作状态影响极大。而一旦滚动轴承发生故障,则会使发动机转子振动不断增大,更严重的话会造成发动机损坏,进而造成严重的飞行事故。因此,对航空发动机滚动轴承进行状态监测和故障诊断,确保其正常工作对发动机甚至飞机的安全可靠地工作是至关重要的。本文采用小波变换对获取到的滚动轴承振动信号进行降噪处理,利用时域和频域分析法提取振动信号的特征属性,构建故障样本集。本文提出了结合SOM和改进后的LE算法进行故障诊断的新模式,设定了军用航空发动机滚动轴承进行故障诊断的一般步骤,并使用该模式首次运用到军用航空发动机故障诊断领域。实验结果表明,该模式可以有效地提高军用航空发动机滚动轴承故障诊断的效率和准确率。
[Abstract]:With the development of science and technology, the reliability requirement of military aero-engine is increasing day by day.However, due to the continuous expansion of the function and the harsh service conditions, the structure of the military aero-engine mechanical parts is becoming more and more complex, and the probability of failure is becoming higher and higher.The causes of failure or failure of military aero-engine parts are complicated and affected by human error, material defects, use environment change, fatigue, wear and aging effects and so on.In this way, it is more and more difficult to diagnose and analyze the malfunction of military aero-engine, and it is necessary to consume a lot of manpower and material resources to ensure the reliability of aircraft.Although the traditional fault diagnosis technology can carry on the fault diagnosis, but the diagnosis efficiency and the accuracy are not high.In order to meet the urgent need in the field of military aero-engine fault diagnosis, this paper applies the theory and technology of data mining, combining neural network and manifold learning algorithm, to the fault diagnosis of military aero-engine.This paper introduces the research status of fault diagnosis technology and data mining technology at home and abroad, as well as the background knowledge needed for the fault diagnosis of rolling bearing of military aero-engine, including the basic vibration mechanism.Common fault types and traditional diagnosis methods, and then the characteristics of rolling bearing vibration signal are described in detail.In this paper, two kinds of classifiers, PNN and SOM, which are commonly used in the field of fault diagnosis by neural network are explained and compared in detail. The experimental results show that the SOM classifier is more effective in fault diagnosis.In this paper, the theory and method of manifold learning are deeply studied, and three common manifold learning algorithms, namely: ISOISOMAPLLE and le, are analyzed.Compared with the traditional linear dimensionality reduction method (PCA) and the nonlinear dimensionality reduction method (LEE), an improved method of le algorithm is proposed. The improved distance function is used to replace the original Euclidean distance function, which can better reflect the local structure information of the high dimensional data set.Improve the performance of le algorithm.This paper mainly studies the problem of fault diagnosis of military aeroengine rolling bearing, which is one of the core parts of military aero-engine, and plays the role of bearing and transferring load.The operating state of the engine has a great influence on the working state of the engine.Once the rolling bearing fails, the vibration of the engine rotor will increase, and the engine will be damaged, and then the serious flight accident will be caused.Therefore, the condition monitoring and fault diagnosis of the aeroengine rolling bearing is very important to the safety and reliability of the engine and even the aircraft.In this paper, wavelet transform is used to reduce the noise of the vibration signal of rolling bearing. The characteristic attributes of vibration signal are extracted by time-domain and frequency-domain analysis, and the fault sample set is constructed.In this paper, a new mode of fault diagnosis based on SOM and improved le algorithm is proposed, and the general steps of fault diagnosis for rolling bearings of military aero-engine are set up.The model is used in the field of military aero engine fault diagnosis for the first time.The experimental results show that the model can effectively improve the efficiency and accuracy of rolling bearing fault diagnosis of military aero-engine.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：V263.6;TP18;TP311.13

【相似文献】