【作者】 苗学问；

【导师】 洪杰；

【作者基本信息】 北京航空航天大学 ， 航空宇航推进理论与工程， 2008， 博士

【摘要】 准确预测主轴承的使用寿命将会对改善航空发动机的可靠性、适用性和安全性起到关键作用。本文针对航空发动机主轴承的工况、使用特点,开创性的采用“状态寿命”(即状态良好、初步损伤、故障发展和即将失效4个寿命阶段)描述主轴承的使用寿命,并建立了基于飞行参数记录数据,确定主轴承状态寿命的模型。该模型由理论计算模型和状态寿命评估模型两部分组成。建立了滚动轴承状态寿命的理论计算模型。针对航空发动机主轴承受载复杂的特点,采用拟动力学方法计算滚动轴承的载荷分布,结合额定动载荷理论修正L-P寿命模型,建立任意载荷下主轴承的寿命计算模型,并研究了不同结构参数和载荷参数对轴承寿命的影响规律;在主轴承的寿命仿真计算、主轴承载荷分析的基础上,结合对飞行参数记录数据典型变化特征的统计分析,以转速和过载为基准,确定主轴承的典型工况和载荷谱,建立主轴承的载荷提取模型。这样根据飞行参数记录数据即可计算主轴承的累积寿命消耗,基于可靠度给出主轴承状态寿命的理论计算值,用以描述主轴承“状态”的统计规律。为了提高理论预测的确定性,建立了以状态监视为依据的滚动轴承状态寿命评估模型,基于振动分析智能评估主轴承个体状态寿命的“经验值”。状态寿命评估模型建模的关键:一是构造状态寿命特征向量;二是状态寿命的辨识算法。结合滚动轴承的全寿命试验,综合考虑诊断能力和灵敏度,提出了基于时域统计量和小波包重构信号的频带能量分布两种构造状态寿命特征向量的方案。由于滚动轴承损伤传播过程的高度非线性、失效模式的多样性和损伤机理的不确定性,本文采用BP神经网络和支持向量机作为状态寿命评估的算法。研究表明,基于贝叶斯正则化方法的BP网络模型的可信性良好;与改进的BP神经网络相比,支持向量机具有鲁棒性、模型训练具有稳定性以及参数优化选择方法具有可移植性等明显的优势,是一种适合工程应用的算法;特征向量的选择显著影响状态寿命评估模型的性能,以小波包频带能量特征向量为输入的评估模型,其收敛速度、准确性均明显优于基于统计量特征向量的模型。最后,基于模糊逻辑推理建立滚动轴承的状态寿命模型,该模型融合状态寿命的理论计算模型和评估模型,确定滚动轴承的状态寿命。设计并建立了滚动轴承多功能试验系统,开展了滚动轴承的全寿命试验,试验验证了本文方法的有效性和可行性。本文的研究为进行航空发动机主轴承使用寿命的预兆和管理提供了一种全面、可靠而独特的方法。更多还原

【Abstract】 Development of practical and verifiable prognostic approaches for gas turbine engine bearings will play a critical role in improving the reliability and safety of aircraft engines. In present study Grade-life was used to describe the bearing’s service life, which means the entire service life was divided into four stages: good bearing condition, initial defect condition, damaged bearing condition and failure coming condition. In addition, a Grade-life prognostic model was presented that utilized available airborne sensor information to calculate and assessment Grade-life for aeroengine bearings. The model comprised of mathematical model and diagnostic estimate model.The mathematical model was a physics-based modeling.In order to set up the formula for bearing’s life calculating, the mechanical models of roller bearings were set up by thequasi-dynamic method, and prognostics equations for bearing’s life calculating was modified based on quasi-dynamic method and basic dynamic capacity theory of bearings. The rules of the life in different structure parameters and load parameters were analyzed, then a load spectra compilation method was presented for aero-engine bearings based on statistical analysis of the flight parameter variety character and aero-engine control schedule. Utilizes information from the Sensed Data module to calculate the cumulative damage sustained by the bearing since it was first installed and the Predition Grade-life (PGL) could be captured based on reliability.Diagnostic estimate model was an“empirical”lifetime model, in which“experience”value of Grade-life (EGL) was assessed based on vibration signals intelligently. Signal feature extraction (Grade-life feature vector) and pattern recognition algorithm of Grade-life were keys to construct the model.Vibration signals of the rolling bearing were analyzed, the selection criterion of the Grade-life feature vectors should include the diagnostic ability, the sensitivity, the consistency and the amount of calculation. Finally, two types Grade-life feature vectors were identified :(1) some time-domain statistical parameters were selected to constructe the feature vector; (2) construction of feature vector by extract feature in every frequency bands by the wavelet packet analysis. Due to the stochastic nature of a failure propagation process, the uncertain mechanical failture model of bearings, the modeling methods of Diagnostic estimate model based on BP neural networks and Support Vector Machines (SVM) were presented and developed in detail. Two examples were choosed to establish the model respectively based on BP network. The result showed that BP network based on Bayesian Regularization method could be successfully used for modeling of EGL. For SVM, the SVM model in establishing the identification model by using bearing test stand run-to-failure data as learning samples was employed. A detailed contrast was performed between the two methods, the result showed the SVM had more evident superiority and it was a technique even more suitable for practical application. In addition, the performance of Diagnostic estimate model was influenced not only by pattern recognition algorithm, but alse by the selection of Grade-life feature vector. For example, extract feature in every frequency bands by the wavelet packet analysis was better than the feature vector of time-domain statistical parameters.The final Grade-life of bearings was determined by fusion of PGL and EGL value by fuzzy logic organon, which reduced the mathematical model uncertainty’s affection towards prediction results of the mathematical model. The validity and creditability of model had been demonstrated by bearing test stand run-to-failure dates.Experimental equipment was constructed to perform accelerated testing on bearings for method verification, in which load coule be put on test bearings by a hydraulic loading mechanism accurately, and the vibration signals were acquired continuously by a sensors and data acquisition sub-system.The Grade-life model provided a scientific, reliable, and unique methodology prognostic the lifetime of aero-engine bearings.更多还原