大型回转支承研究——博士论文.docx

1、博士学位论文I摘 要随着工程机械和风力发电行业深入地发展，相关产业已经从增量市场逐步向存量市场过渡，保证此类大型设备的可靠运行逐步成为企业的工作重点。作为在这两个行业中广泛使用的重要回转连接件大型回转支承，其故障或失效引发的事故屡见不鲜，造成了严重的社会影响和经济损失。究其原因，多是有效维护的缺失。要实现设备的有效维护，就必须能够评估设备的实时健康状态，预测设备的健康发展趋势，这已成为相关企业亟需解决的关键问题之一。设备的“健康 ”可定义为设备是否出现故障、性能退化的程度、剩余的使用寿命等。近年来，设备的在线健康监测技术得到了极大的关注，它能够准确地评估和预测设备的健康状态，从而在设备出现异常

2、前采取相应的维护措施，避免重大事故的发生。研究设备的在线健康监测，实质上是研究设备的故障诊断方法，揭示设备的性能退化规律，掌握设备的寿命预测理论，最终建立起设备的在线健康监测系统，为设备的有效维护提供可靠的理论和技术支撑。这对于企业制定有效的主动维护策略，降低设备故障风险，提升在行业中的竞争力有着重要的意义。为此，本文以大型回转支承为研究对象，深入开展了其在线健康监测理论与应用的研究，包括以下几方面的内容：（1）提出了回转支承的小样本加速寿命试验方法，建立了回转支承疲劳寿命的威布尔分布模型，进而实现了任意工况载荷下回转支承的可靠性寿命预测。其中，小样本试验方法依据滚道载荷分布规律和 Archa

3、rd 磨损理论，通过一次加速寿命试验即可建立滚道载荷磨损量伪失效疲劳寿命的关系，进而获取到多个应力水平下多个回转支承滚道的失效样本，大幅降低了试验成本。利用这些样本建立起威布尔分布模型后，再运用逆幂率加速模型实现了任意载荷下回转支承寿命的可靠性预测。研究发现，在类似风机变桨回转支承的实际载荷工况下，回转支承定圈滚道损伤程度较高，滚道磨损量与各区域接触载荷密切相关。相比 ISO281 和 NREL DG03 中的 L10 寿命，本方法预测结果更接近工程实际，为回转支承的设计优化提供了指导依据。（2）针对现有降噪方法的不足，提出了一种基于 EEMD-KPCA 的全寿命振动信号降噪方法。该方法首先利

4、用 EEMD 将全寿命周期中不同时期的振动信号分解成Error! No text of specified style in document.II多个 IMF，然后利用 KPCA 中的 SPE 指标评价各 IMF 在不同时期的差异，从而选取在整个寿命周期中最能反映回转支承性能退化过程的多个 IMFs 进行信号重构，完成信号的降噪。此外，深入研究了 EEMD、WT 阈值法、KPCA 主元法等常用降噪方法中的参数优化理论和降噪过程。对比分析发现，提出的 EEMD-KPCA 方法比常用降噪效果更好，而且降噪信号可解释性强，保留了足够多的回转支承性能退化信息，为后续研究提供了可靠的预处理手段。（3）

5、提出了基于圆域分析的回转支承故障诊断方法。该方法首先探讨了 PAA过程中的参数优化，然后利用振动信号邻域相关离散点拟合椭圆的倾角方向对信号中频率变化的敏感性，来识别信号中存在的异常信息，进而将其应用到回转支承整圈滚道中实现其初期的故障诊断。相比常见方法，该方法并不关注信号中具体的特征频率，而是以圆域特征反映出回转支承整体的异常情况，克服了回转支承振动信号中的有效频率由于能量过低而难以提取的问题。结果表明：相比时域特征和小波分析，圆域故障诊断方法能够更为准确地诊断出回转支承的初期故障，且故障点处的可识别度更高。（4）提出以 C-SPE 指标评估大型回转支承的性能退化过程，并据此建立了基于状态数据

6、和信息融合的 RUL 预测模型，形成了 RUL 预测理论体系。首先，利用振动信号的时域特征揭示了大型回转支承的性能退化过程；然后，提出以 KPCA 中的统计量 SPE 来量化分析多维振动信号的差异，并利用 C-SPE 评估回转支承性能退化过程。此外，深入讨论了 LS-SVM 的 GSA-PSO 参数优化方法，建立起基于状态数据的 RUL 预测模型；最后，利用失效率修正了威布尔分布的 RUL 预测模型，提出了一种结合可靠性和状态数据的信息融合 RUL 预测模型。研究表明：多维振动信号能够提供更全面的回转支承退化信息，而 C-SPE 能够在保留足够多的退化信息的同时大幅降低特征维数；GSA-PSO

7、 比 CV、GA、PSO 等方法具有更高的建模效率和精度；基于 C-SPE 的 RUL 预测模型比基于时域特征的模型更优，且温度、驱动力矩等辅助参数有利于进一步提高模型精度；基于实时失效率的信息融合 RUL 预测模型计算效率较高，其结果更接近工程实际，可以用于回转支承在线健康监测系统。（5）针对现有监测系统的不足，利用 NI 的软硬件平台研发了大型回转支承的在线健康监测系统。该系统设计了多路同步采集、基于 C-S 架构的 TCP 通讯协议、友好的人机界面、安全的数据管理等功能。此外，通过将 Matlab 算法移植到cRIO，该系统实现了回转支承振动信号采集与降噪、故障诊断、特征提取、失效率博士

8、学位论文III评估、剩余寿命预测等在线健康评估功能的分布式计算，在多个风场的风机监测中得到了应用。关键词：大型回转支承 信号降噪 故障诊断 性能退化评估 寿命预测 在线健康监测博士学位论文iABSTRACTIn recent years, with the rapid development of construction machinery and wind turbines, such industries have been changing from incremental market to stock market. Ensuring reliable operation of t

9、hese large equipment is becoming more and more important for relevant enterprises. As a vital rolling rotational connection, the fault or complete failure of a large size slewing bearing may cause some serious incidents, which brings great losses in both social and financial aspects. Therefore, how

10、to make proactive maintenance strategy to increase the reliability of slewing bearings turns into one of the critical issues.Generally speaking, online health assessment and prediction is the prerequisite to achieve proactive maintenance, which make it possible to repair or replace some components w

11、hen incipient abnormality appears. In order to help enterprises increase their competitiveness, it is required to research diagnosis methods, reveal performance degradation patterns, study life distribution characteristics, and develop online health monitoring systems (HMS), which provides reliable

12、theories and technical supports for proactive maintenance. Hence, with the funding of national natural science foundation and provincial foundations, this paper focuses on residual useful life (RUL) prediction theories and online HMS of large size slewing bearings, of which the contents are as follo

13、ws:(1) A small sample accelerated life test approach is proposed to build the Weibull distribution model of the slewing bearing fatigue lives, which further achieves reliability-based RUL prediction under any load cases. Based on the load distribution of the raceway and Archard theory, the small sam

14、ple test approach can obtain multiple failure samples under multiple load conditions by building the relationship between load levels, wear volume and pseudo fatigue life, which effectively decreases the test cost. Afterwards, the Weibull model can be established and the reliability based RUL can be

15、 predicted using Anti-power theory. Experimental results show that the fixed ring raceway of the slewing bearing damages earlier and more severe than the turntable ring under a load case in wind turbine pitch system, and the wear volume of each raceway section is strongly related to Error! No text o

16、f specified style in document.iithe load level. Besides, compared to the L10 life in ISO 281 and NREL DG03, the results of the proposed method is more close to engineering practice, and thus providing a reference for slewing bearing design.(2) A novel EEMD-KPCA de-noising method is developed based o

17、n life-cycle vibration signals to overcome the shortcomings of commonly used methods. Firstly, the proposed method decomposes vibration signals obtained from different time periods into several IMFs using EEMD. Afterwards, SPE statistics of KPCA is employed to evaluate the differences of a same IMF

18、in different time periods. Finally, signal de-noising can be achieved by reconstructing the IMFs that are capable of revealing the whole performance degradation processes of the slewing bearing. In addition, the parameters optimization and de-noising of commonly used methods such as EEMD, WT and KPC

19、A are also studied. Comparative results show that the proposed EEMD-KPCA has a better de-noising performance, and the de-noised signal contains sufficient degradation information of the slewing bearing, which provides a reliable pre-processing approach for further research.(3) A circular domain anal

20、ysis method is proposed for slewing bearing diagnosis. First of all, the time domain signal is transformed into circular domain and divided into several zones according to a certain angle, and then the neighborhood correlation discrete points of each angle zone are fitted as an ellipse. Afterwards,

21、the ellipses that skews to the right are tagged as abnormality and the corresponding abnormal vectors are obtained based on the whole cycle of the slewing bearing. Finally, the mean vector of all the abnormal vectors is acquired, and the mean, variance, skewness and kurtosis of the character vector

22、are regarded as the fault indicators. Compared with current methods, the proposed method focuses on the abnormality of the whole slewing bearing while not the specific character frequency, which overcomes the shortcoming that the effective frequencies are difficult to extract from low SNR slewing be

23、aring vibration signals. Results show that the proposed method is able to diagnose the incipient fault of a slewing bearing more accurately and identifiably than time domain features and wavelet analysis.(4) C-SPE and its time-domain features are introduced as the performance degradation indicators

24、of the slewing bearing. On this basis, RUL prediction system is established by proposing the condition data driven and data fusion methods, together with 博士学位论文iiithe previous proposed reliability based method. To begin with, time domain features of multi-dimensional vibration signals are used to re

25、veal the performance degradation process of the slewing bearing. Then, SPE statics of KPCA is employed to quantitatively analyze the difference of multi-dimensional signals in different time periods, and C-SPE is thus developed to describe the performance degradation process. Afterwards, GSA-PSO opt

26、imization method is proposed and studied to determine the best parameters of LS-SVM, and then condition data driven RUL prediction model is built based on condition monitoring data. Last but not least, the Weibull RUL prediction model is modified based on failure rate, which makes it possible to pro

27、pose a data fusion RUL prediction method that combines reliability and condition monitoring data. Results show that multi-dimensional vibration signals can provide more comprehensive degradation information, while C-SPE can reduce the dimensionality of features at the same time. GSA-PSO has a higher

28、 efficiency and accuracy than commonly used CV, GA and PSO methods. C-SPE based RUL prediction model is better than time domain features based model, and temperature and driven torque may help improve the accuracy of the RUL prediction model. Above all, the results of the proposed data fusion method

29、 are more close to engineering practice by taking both load and real-time condition into consideration, which makes it suitable for online health monitoring of slewing bearings.(5) An online health monitoring system (HMS) for slewing bearings is developed using NI hardware and software platform. The

30、 HMS provides multiple functions such as multi-channel synchronous data acquisition, C-S structure based TCP communication protocols, friendly human machine interface, secured data management and so forth. More importantly, by transplanting Matlab algorithms to cRIO platform, the HMS is able to acco

31、mplish signal de-noising, fault diagnosis, feature extraction, failure rate evaluation and RUL prediction remotely, which provides a potential for wind turbine monitoring in wind power plants.KEYWORDS: Large-size slewing bearing; Signal de-noising; Fault diagnosis; Performance degradation assessment

32、; RUL prediction; Online health monitoringError! No text of specified style in document.iv博士学位论文目 录摘 要 .IABSTRACT .i第 1 章 绪论 .11.1 选题背景和研究意义 .11.2 国内外研究现状 .31.2.1 振动信号降噪与故障诊断方法 .31.2.2 性能退化评估与寿命预测模型 .61.2.3 回转支承在线健康监测系统 .141.3 课题来源 .151.4 主要研究工作与技术路线 .16第 2 章 基于小样本加速寿命试验的寿命预测可靠性模型 .192.1 引言 .192.2 威

33、布尔分布理论及参数估计 .192.2.1 基本理论及剩余寿命可靠性模型 .192.2.2 模型参数估计方法 .212.3 大型回转支承小样本加速寿命试验方法 .252.3.1 总体流程 .252.3.2 滚道接触载荷分布及分段 .262.3.3 磨损量与伪失效寿命关系模型 .312.3.4 滚道最大载荷与特征寿命关系模型 .332.4 大型回转支承全寿命试验 .342.4.1 试验装备简介 .342.4.2 数据采集系统 .352.4.3 试验方案与结果 .36Error! No text of specified style in document.2.5 试验验证与对比研究 .402.6

34、本章小结 .45第 3 章 大型回转支承非平稳振动信号的降噪方法 .473.1 引言 .473.2 集合经验模态分解(EEMD)降噪方法 .473.2.1 基本原理及降噪 .473.2.2 参数对降噪效果的影响 .503.3 核主元分析(KPCA)理论及应用 .543.3.1 基本原理 .543.3.2 参数优化及降噪 .563.3.3 KPCA 用于异常识别 .583.4 小波变换(WT)降噪方法 .603.4.1 基本原理 .603.4.2 参数对降噪效果的影响 .623.5 基于全寿命振动信号的 EEMD-KPCA 降噪方法 .643.6 试验验证与对比研究 .673.7 本章小结 .72第 4 章 大型回转支承圆域分析故障诊断方法 .734.1 引言 .734.2 基本理论 .734.2.1 PAA 与邻域相关图 .734.2.2 离散点椭圆拟合法 .754.2.3 圆域重采样 .774.3 基于圆域分析的故障诊断 .794.3.1 故障诊断流程 .794.3.2 PAA 窗优化方法 .804.3.3 圆域特征计算 .824.4 试验验证与对比研究 .834.5 本章小结 .88