【作者】 郭磊；

【导师】 陈进；

【作者基本信息】 上海交通大学 ， 机械设计及理论， 2009， 博士

【摘要】 随着科学技术的进步和工业需求的发展,各类先进生产设备一方面不断向复杂、高速、高效、轻型、微型或大型的方向发展,另一方面却又面临更加苛刻的工作和运行环境。一旦设备的关键部件发生故障,就可能破坏整台设备甚至影响整个生产过程,造成巨大经济损失,还可能导致灾难性的人员伤亡并产生严重的社会影响。因此,如何有效评估设备的运行状态,从而能够及时采取措施以防止灾难性事故的发生是当前迫切需要解决的问题。设备从性能开始恶化到完全失效,要经过一个性能逐步退化的过程。如果能够在设备性能退化的过程中检测到设备性能退化的程度,那么就可以有针对地组织生产和设备维修,防止设备异常失效的发生。设备性能退化评估与预测正是基于以上思想提出的一种主动设备维护的技术。设备性能退化评估与预测侧重于对设备性能衰退状态全过程的走向预测,而不在于某个时间点的性能状态诊断,因此,其与现有的故障诊断技术在理念上和方法上都有很大的不同。本文以旋转机械的典型零部件为对象,深入开展了设备性能退化智能评估和预测的理论体系和技术方法研究,包括以下几个方面的内容:从理论分析与工程应用的角度出发,阐述了论文的选题背景和研究意义。分析了设备性能退化评估方法、预测方法以及核模式分析方法等方面的国内外发展现状,总结了目前研究中需要解决的问题,确立了本论文的研究内容。简要地叙述了核模式分析方法的原理,将小波核函数引入核模式分析方法中,研究了小波核函数在支持向量机分类器、支持向量机回归分析以及核主元分析中的应用。基于平移不变核函数Mercer条件,推导证明了Mexican hat小波函数构造的容许小波核函数。利用仿真数据和轴承试验数据,对比了小波核模式分析方法与RBF核模式分析方法的泛化能力,分析结果表明前者具有更好的性能。提出了循环平稳熵分析方法,通过对轴承振动信号的相关性进行定量分析,准确地监测轴承运行状态。此外利用小波核函数核主元分析方法对多维特征向量进行约简,在全面掌握设备性能的同时,提高分析的效率并增强评估的准确性。利用轴承的试验数据对特征提取和特征约简方法进行了验证。研究了基于支持向量机的多分类器性能退化评估方法。本文应用支持向量机二叉树算法进行设备性能退化评估研究,避免了传统支持向量机多分类器的拒识区域问题。将小波核函数引入支持向量机多分类器中,提高了分类器的分类精度。提出了几何距离概率统计准则,在保证分类精度的同时,提高支持向量机多分类器参数优选的效率。使用不同故障程度的轴承试验数据对核参数优化准则及分类算法进行了验证。研究了基于支持向量机几何距离的性能退化评估模型。支持向量机算法在处理二分类问题时,首先利用核函数将特征向量映射到高维特征空间,然后在该特种空间中建立线性分类面进行分类。本文提出的评估方法利用设备状态特征向量与支持向量机最优分类面之间的几何距离,描述设备偏离正常状态的程度,从而实现对设备性能退化程度的定量分析。利用裂纹转子仿真模型,对评估模型的泛化推广能力进行了分析。根据拉依达准则,研究了连续变量的自适应报警阈值设定问题。裂纹转子和轴承疲劳试验数据验证了该评估方法的评估效果。研究了基于小波核支持向量机自回归(WSVAR)模型的性能退化预测方法。研究了模型各个参数对预测结果的影响,并以二进制网格搜索方法选定模型最优参数。利用轴承加速疲劳试验全过程数据,对于WSVAR模型的预测效果进行了验证。并将WSVAR的预测结果分别与RBF核函数SVAR和RBF神经网络的预测结果进行了对比,表明WSVAR预测模型具有更高的预测精度。进行了轴承强化疲劳实验研究。在轴承加速疲劳试验台上采集了多组轴承强化疲劳试验全过程数据,利用大量的滚动轴承信号验证了文中提出的各种特征提取方法、特征约简方法、性能退化评估和预测方法的适用性。针对机械设备功能层次分级且相对独立的特点,提出了基于层次分析的设备整体性能退化评估系统模型。考虑系统计算量大及系统后续扩展性等问题,在系统的开发中引入了面向服务(SOA)技术,并基于SOA的成熟实现技术WCF开发了设备性能退化评估与预测系统原型。在某风机监测系统的现场应用验证了系统设计的有效性。更多还原

【Abstract】 The rapid growth of technologies and market competition has already had a significant impact on commercial manufacture. The equipments development direction has some new characteristics, such as hugeness, distribution, high speed, automation and complexity. And also the equipments must face more and more harsh running condition. Once there is something wrong with them, production efficiency will fall or machine sets halt, even catastrophic accidents will occur. It is necessary and important to monitor key equipments and diagnose their faults in order to improve safety, allow predictive maintenance and shorten significantly the associated out of service time.Equipment performance degradation is a continuous process, and there are several stages from the initial degradation to the final failure. If the degree of the equipment performance degradation can be detected, it would be possible to make credible maintenance schedule and prevent the urgent broken. Equipment performance degradation assessment and prediction is proposed based on the above idea. Unlike the traditional fault diagnosis research, whose aim is to fault type classification, the emphases of this research is the equipment degradation degree and trend. By taking rolling element bearing and rotor as the research basis, several degradation assessment and prediction methods are brought. The contents are as follows:From the viewpoint of theoretical analysis and engineering application, the background and significance of the present study are elucidated. The state of the art review on equipment degradation assessment and prediction technology is thoroughly completed, respectively. The concrete research points are decided, then the research content of this paper are defined.The principles of kernel analysis method are briefly talked about. The wavelet kernel is inducted to several kernel analysis methods, such as support vector machine, support vector regression and kernel principle component analysis. Based on the translation invariant kernel Mercer theorem, the admissible wavelet kernel is constructed based on Mexican hat mother wavelet. Utilizing emulation and testing data set, the generalization ability of kernel analysis method using wavelet kernel and the one using RBF kernel are compared. The results indicate that the former has better performance.The definition of cyclostationarity entropy is brought forward as a monitoring tool according to CS characters of rolling element bearing. cyclostationarity entropy reflects the correlation between spectral lines, which will change with the development of failure and the deterioration of machine’s operation situation. Wavelet kernel principle analysis (WKPCA) method is used to reduce the dimension of characteristic vector. WKPCA can improve the assessment efficiency and veracity. By using testing bearing data set, the performance of cyclostationarity entropy feature extraction method and WKPCA feature reducing method are verified.The degradation assessment method based on SVM multi-classifier is proposed. By utilizing the SVM binary tree algorithm, the decline problem of SVM based on“one-against-one”or“one-against-other”strategy is solved. The classification accuracy is improved by using the wavelet kernel. In order to increase the SVM training efficiency, the geometric distance probability parameters optimization is proposed. By using bearing data set with different pitting diameter, the performance of SVM assessment method and parameters optimization method are verified.The degradation assessment method based on SVM geometric distance is proposed. During solving the binary classification problem, SVM algorithm will construct an optimal hyperplan. The geometric distance between the vector and the hyperplan is taken as the measurement of degradation degree. Using the cracked rotor emulation data sets, the generalization of this assessment method is studied. Based on Pauta Criterion, the self-adapting alarm technology is studied.The performance degradation prediction method based on wavelet kernel support vector auto regression (WSVAR) is proposed. The inferences of different parameters on the prediction result are discussed. By using the binary grid searching method, the optimum prediction parameters are selected. By using bearing accelerated life testing data set, the prediction results of WSVAR and RBF-ANN and RBF-SVAR are compared. The results indicate that the WSVAR has more prediction accuracy.A bearing accelerated life test is performed on the accelerated bearing life tester (ABLT-1A) and several bearing data sets are collected. The effective of these above feature extraction method, the performance degradation assessment method and degradation prediction method are validated.A whole equipment performance degradation assessment system model based on analytic hierarchy process is proposed. And the system is developed based on the distributed programming model, WCF (Windows Communication Foundation), which is the realization technology of SOA (Service-Oriented Architecture). The utilization of WCF can solve the problem that the realization of performance degradation assessment algorithm is too complex to be satisfied with a single computer. A gas blower performance degradation assessment system is realized on the .NET platform, which validates the availability of the design.更多还原