CASIA OpenIR  > 毕业生  > 硕士学位论文
基于非线性时间序列分析的航空发动机健康管理预测技术研究
其他题名Technical Research on Prognostics of Aircraft Engine Health Management Based on Nonlinear Time Series Analysis
吕骘
学位类型工程硕士
导师王健
2015-05-28
学位授予单位中国科学院大学
学位授予地点中国科学院自动化研究所
学位专业控制工程
关键词航空发动机 时间序列分析 预测 过程神经网络 相似性 Aircraft Engine Time Series Analysis Prognostics Pnn Similarity
摘要在飞行过程中,作为飞机心脏的航空发动机如果发生故障,将直接威胁到飞机的飞行安全。而如果能够提前知晓发动机发生故障的原因,并有针对性地进行维护、维修或换发,将大大降低飞机失事的可能性。此外,如果能够最小化发动机在其寿命期的维修费用,对航空公司来说是一件极具经济效益的事情。正是在这样的需求背景下,各航空强国及航空公司先后提出了发动机预测与健康管理(PHM,Prognostics and Health Management)的概念。 实际上,预测是健康管理的前提,因此,可以认为预测技术是健康管理技术的一部分。目前,全球最为著名的三家航空制造商(普惠公司、通用公司和罗罗公司)分别推出了一款甚至几款航空发动机状态监视系统。虽然这些系统能够监测、记录、诊断和评估发动机的健康状态,具备一部分的健康管理功能,但它们不具备预测能力。这与当前关于航空发动机预测理论的研究不足有关,有必要进行深入研究。 航空发动机是一个复杂的、随时间变化的非线性系统,所以非线性时间序列分析理论对航空发动机是适用的;而且基于该理论的预测模型已在金融、管理和生物学等领域得到了成功应用,这也为非线性时间序列分析理论在航空发动机健康管理预测技术中的应用提供了丰富的经验。因此,本文围绕非线性时间序列分析理论,分别从航空发动机性能预测、故障预测和寿命预测三方面展开研究。 针对性能预测,本文首先提出基于小波变换和ARIMA模型(Auto-Regressive Integrating Moving Average Model)的方法,并以发动机EGTM趋势预测为例进行了仿真实现。受益于小波变换对信号的分解作用,该方法与基于单一ARIMA模型的预测方法相比,具有更高的预测精度。然而,小波变换对于信号早期的微弱变化无能为力,且ARIMA模型不具备时间累积效应的处理能力,于是又提出基于经验模态分解和离散过程神经网络的方法,并针对同一组实验数据进行了仿真实现。结果表明,基于该方法建立的预测模型具有比基于小波变换和ARIMA模型更优的性能。 针对故障预测,本文首先分析了航空发动机的典型故障模式,并给出故障预测的一般过程。以磨损故障为例,将滑油系统Fe元素含量数据看做混沌序列,提出并实现了基于支持向量机的磨损趋势预测,并与人工神经网络和灰色模型的预测结果作对比,验证该方法的有效性。 针对寿命预测,本文从实验数据入手,以功能失效曲线为基础,提出并实现了基于相似性的预测方法,预测结果表明,该方法达到了预测航空发动机的剩余使用寿命的目的。 上述实验结果表明,基于非线性时间序列分析的航空发动机健康管理预测技术具有实用价值。
其他摘要During the flight, if a fault occurs on the engine which works as the aircraft heart, it will be a direct threat to the flight safety. But if we can know the cause of the fault in advance, and then carry out specific maintenance or replacement, it will greatly reduce the possibility of the aircraft crash. In addition, if the engine maintenance costs can be minimized in its life cycle, the airlines will obtain great benefits. Just under such a background, the air power and airlines have proposed the concept of Prognostics and health management (PHM) successively. In fact, prognostics is the premise of health management, therefore, prediction technology is a part of health management techonology. Currently, the most famous aviation manufacturers (P&W, GE and Rolls-Royce) were separately launched one or several engine condition monitoring system or systems. Although, these systems can monitor, record, diagnosis and evaluate health condition, but they do not have the ability to predict. It is related to the lack of aircraft engine prediiton theoretical research, so it is necessary to conduct in-depth study on it. An aircraft engine is a complex and changing with time nonlinear system, so the nonlinear time series analysis theory is applied to it. What is more, prediction models based ont the theory have been successfully applied in the areas of finance, management and biology and others, which provides a wealth of experience for the application in engine health management. Therefore, this paper focuses on the nonlinear time series analysis theory, and make research on three aspects: eingine performance prediction, fault prediction and life prediction. For performance prediction, a method based on Wavelet Transform and ARIMA models is proposed. Benifiting from the wavelet decomposition of signals, this method presents a higher prediction accuracy compared with the traditional methods of time series analysis. However, wavelet transform has no responds to the subtle small changes in the signals of early processing, and ARIMA models don’t have ability to handle the time cumulative effect. Thus, a new method based on EMD and PNN is proposed. As the experiment shows, this method performances better. For the engine fault prediction problem, some typical fault modes of engines are produced, and then this paper presents the processing to predict engine faults. Because there are a large amount of fault modes, wearing fault is chosed to verify th...
其他标识符2012E8014661086
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/7757
专题毕业生_硕士学位论文
推荐引用方式
GB/T 7714
吕骘. 基于非线性时间序列分析的航空发动机健康管理预测技术研究[D]. 中国科学院自动化研究所. 中国科学院大学,2015.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_2012E801466108(1862KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[吕骘]的文章
百度学术
百度学术中相似的文章
[吕骘]的文章
必应学术
必应学术中相似的文章
[吕骘]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。