【作者】 曾庆虎；

【导师】 邱静；

【作者基本信息】 国防科学技术大学 ， 机械工程， 2010， 博士

【摘要】 机械传动系统作为国防和国民经济领域广泛应用的一类重要技术装备,其安全性可靠性至关重要。齿轮与轴承等传动系统关键零部件,由于长期连续工作在高载荷、高转速下,容易受到损害和出现故障,其损坏往往会导致传动系统无法运转。研究实用、可靠的传动系统关键零部件故障预测技术,实现主动的故障预测,是预防故障、保持机械传动系统战备完好性的技术基础,其研究意义重大。论文以“十一五”部委级预研课题“装备动力传动系统状态实时监控与故障预测技术”为背景,针对机械传动系统关键零部件存在故障演化规律分析建模和故障预测方法的不足,系统分析了机械传动系统关键零部件主要故障机理与故障演化规律,对其故障演化规律进行建模分析研究；在此基础上深入研究了小波相关特征尺度熵特征信息提取技术和动力传动系统关键零部件HSMM退化状态识别与故障预测技术。研究成果对于提高机械传动零部件故障预测能力具有重要的参考价值和指导意义。论文的主要研究内容包括：1.机械传动系统关键零部件故障机理分析与建模在系统地分析机械传动系统关键零部件齿轮与轴承的主要故障模式和失效机理及其故障演化规律基础上,利用故障演化过程中退化状态与HMM都是通过观测值来感知其状态的共同特点和HSMM能合理的描述故障演化过程退化状态驻留时间这一特性,建立了故障演化规律HSMM模型,为机械传动系统关键零部件故障预测技术的研究奠定了基础。2.机械传动系统关键零部件故障预测技术研究(1)为解决退化状态识别与故障预测的特征信息提取问题,研究了小波相关特征尺度熵特征信息提取技术。特征信息提取直接关系到退化状态识别的准确性和故障预测的可靠性,而噪声是影响特征信息提取的最主要障碍,论文基于小波熵理论的基本思想,引入小波相关滤波降噪方法,并将该方法与Shannon信息熵原理相结合,定义了一种新的小波熵概念一小波相关特征尺度熵,进而提出了一种新的用于状态识别与预测的特征信息提取方法一小波相关特征尺度熵特征信息提取方法。研究表明该方法较一般小波熵特征提取方法更能有效、综合的表征设备运行状态,为设备退化状态识别与故障预测的特征信息提取提供了一种新的有效途径。(2)为确定设备当前所处的退化状态以及预测设备的剩余使用寿命,研究了机械传动系统关键零部件HSMM退化状态识别与故障预测技术。首先深入研究了HSMM应用于退化状态识别与故障预测时存在如何恰当选择模型初始参数、下溢和模型泛化等问题,提出了一种基于HSMM状态识别与故障预测的模块化训练算法。基于此研究了HSMM在机械传动系统关键零部件退化状态识别与故障预测中的应用方法,通过实例验证了该方法的可行性和有效性。进一步地,为提高状态识别与预测的精度,充分利用多类传感信息,从多源信息融合的角度出发,将KPCA方法引入,进行多通道特征信息融合,基于融合后新的特征信息,研究了基于KPCA-HSMM的退化状态识别与故障预测方法。研究结果表明：该方法可有效融合状态识别与故障预测中的多源信息,提高状态识别与故障预测的可靠度与准确性。3试验验证以机械传动系统关键零部件滚动轴承为试验研究对象,在杭州轴承试验研究中心有限公司ABLT-7型轴承试验机上进行了滚动轴承全寿命试验研究,用实测全寿命数据验证了本文所研究方法的可行性和有效性。更多还原

【Abstract】 The safety and reliability of the mechanical transmission systems are vital important, which are widely used in the national defense and national economy realms as a type of important technical equipments. The key parts and components of mechanical transmission systems such as gear and bearing are easily suffered damage and fault due to lasting and continuous running under heavy load with high rotate-speed, and hence resulting in transmission system broken. So it is of great significance to study the practical and reliable fault prognostics technologies of the key parts and components of mechanical transmission systems and active fault prognostics realization methods, which are the technical base of the fault prevention and mechanical transmission systems operational readiness keeping.Funded by the advanced project "Equipment Power and Transmission Systems Condition Monitoring and Fault Prognostics Technology", the dissertation systematically analyzes the fault mechanism and fault evolution rules and studies the fault evolution rules modeling of the key parts and components of mechanical transmission systems mainly aiming at the shortage of fault evolution rules analysis modeling and fault prognostics methods. Then, the feature information extraction technology based on wavelet correlation feature scale entropy, as well as HSMM degradation state recognition and fault prognostics technologies are deeply studied. The research findings have great referrence value and guiding meaning on the improvement of fault prognostics ability of the mechanical transmission parts and components.The main contents of the dissertation are as follows.1. Fault mechanism analysis and modeling of the key parts and components of mechanical transmission systemsOn the base of systematical analysis of the main fault mode, invalid mechanism and fault evolvement rules of mechanical transmission systems key parts and components gear and bearings, the fault evolution rules HSMM model is constructed making use of the similarity between the degradation state in the fault evolution process and HMM (both are apperceived by the observations) and the characteristic that HSMM can logically describe degradation state resident time in the fault evolution process, which lay a solid foundation for the research of fault prognostics technology of the key parts and components of mechanical transmission systems.2. Fault prognostics technology studies of the key parts and components of mechanical transmission systems(1) The feature information extraction technology based on wavelet correlation feature scale entropy is studied in order to resolve the feature information extraction problem of the degradation state recognition and fault prognostics. Feature information extraction has direct relation to the accuracy of the degradation state recognition and the reliability of the fault prognostics, yet the noise is the main obstacle. So the dissertation presents a new feature information extraction method, namely the wavelet correlation feature scale entropy-based feature information extraction technology, which is based on the basic idea of the wavelet entropy theory and conjuncts with wavelet correlation filter denoise method and Shannon information entropy principle. The proposed method can denote equipment operation state more efficiently and synthetically compared to the normal wavelet entropy-based feature extraction method, and hence provides a new effective way for the feature information extraction of the degradation state recognition and fault prognostics.(2) HSMM degradation state recognition and fault prognostics technologies are studied in order to confirm the current degradation state and prognosticate the remaining useful life of the equipments. Firstly, the problems, when HSMM is used to the degradation state recognition and fault prognostics domain, such as how to select model initial parameter, underflow and model generalization and so on are studied, and a modularized training algorithm based on HSMM state recognition and fault prognostics is proposed. Then, the application method about how the HSMM is use to the degradation state recognition and fault prognostics of the key components of power and transmission systems is studied and its feasibility and validity were proved by demonstrations. Further more, in order to improve state recognition and prognostics precision and make full use of multi-source sensor information, the KPCA-HSMM degradation state recognition and fault prognostics method is studied, which involves the KPCA so as to fuse multi-source feature information. The study outcomes indicate that the proposed method can fuse the multi-source information in the state recognition and fault prognostics domains and can improve the reliability and accuracy of the state recognition and fault prognostics.3. Experiment verificaitonThe rolling bearing, which is a typical key parts and components of mechanical transmission systems, is selected as the study object. The experiment was carried out on the ABLT-7 bearing testing machine in the Hang Zhou Bearings Testing and Research Center Ltd, and the tested life cycle experimental data verified the feasibility and validity of the proposed method.更多还原