滚动轴承服役期间的性能可靠性对于工作主机的运行势态有着重要的影响。由于应用过程中的安装条件、应力波动性、润滑条件的不确定性以及轴承本身所携带的特殊信息等,滚动轴承的性能演变与失效概率分布信息不明确,难以采取以应力-强度干涉理论为基础的疲劳寿命概率设计的经典统计方法预测其性能可靠性;虽然服役期间外部环境与内部因素的复杂性与多变性导致轴承性能可靠性在未来的演变动态存在不确定性,但正常运行轴承的某一时间序列的动态性能能够提供轴承性能的演变信息。 开展专门试验,为可靠性演变过程预测及试验验证提供数据支持。性能时间序列是开展性能数据驱动的滚动轴承可靠性演变过程预测方法研究的前提和基础,考虑到不同结构类型的滚动轴承以及实际应用中的不同阶段对于性能参数值的影响,针对汽车水泵轴连轴承与风力发电机主轴用双列圆锥滚子轴承分别开展工况模拟台架试验与实际运行性能监测试验,获取定时截尾、依据性能指标的定数截尾以及完全寿命周期试验的滚动轴承性能时间序列。 采用非线性动力学分析和时间序列混沌特征识别两种识别方法,从不同角度确定滚动轴承性能的混沌特性。选取滚动轴承内部的典型元件,构建滚动轴承运行中的非线性动力学方程,设置运行条件,采用数值方法求解非线性动力学方程,识别出滚动轴承性能的混沌特性;重构测量获取的性能原始时间序列的相空间,计算性能时间序列的混沌特征参数,验证滚动轴承性能的混沌特性。 预测轴承性能的未来时间序列,获取轴承可能的性能寿命值,建立滚动轴承的性能可靠性预测模型。由于系统信息或数据的不完备性,单一数学方法预测得到的结果难以反映轴承性能演变状态的全面信息,采用15种混沌预测方法预测未来时间的性能轨迹,获取1 5个未来性能时间序列;依据工况需求确认失效值,划定性能失效域,得到不同预测方法对应的性能发展趋势曲线;求取失效值与性能发展趋势曲线的交点,获得多个可能的轴承性能寿命值;采用灰白助法对预测数据开展灰自助,获取充足数据构建直方图,建立滚动轴承性能可靠性的概率密度函数,实现数据驱动的轴承性能可靠性预测。 通过概率密度函数的演变过程揭示滚动轴承性能可靠性的演变状态。基于最大熵原理建立最大熵条件下滚动轴承性能的概率密度函数,引入Lagrange乘子作为概率密度函数的特征参数;将原始时间序列划分为多个子时间序列,利用不同子序列的预测寿命数据得到携带不同Lagrange乘子信息的滚动轴承性能的概率密度函数,由概率密度函数的变化可揭示滚动轴承性能可靠性的演变;采用层级挖掘方法与分段挖掘方法来挖掘概率密度函数演变过程中的不同侧面信息,为提取概率密度函数特征参数的一致性信息,采用信息融合方法获取综合多种侧面信息的滚动轴承性能可靠性的概率密度函数;利用层级挖掘与分段挖掘获取的Lagrange乘子特征值集合,获取Lagrange乘子的估计真值与置信区间上、下限,组合Lagrange乘子集合的上、下限寻找性能可靠性概率密度函数的上、下限,实现滚动轴承性能可靠性的评估。 开展滚动轴承性能可靠性的假设检验,利用假设检验的双驱统计量发掘滚动轴承性能可靠性背离点,建立性能可靠性演变的不确定性概率模型,揭示可靠性演变的不确定性机制。利用模糊等价关系概念提取滚动轴承性能可靠性预测效果的等价关系发生概率,利用Bayes统计学建立后验发生概率,获取假设检验否定域;两种假设检验方法形成双驱统计量,依据双驱统计量的量化指标发掘轴承性能可靠性的变异点;利用动力学理论,分别从速度和加速度两个不同角度研究滚动轴承性能可靠性的演变过程,确定其不确定性的随机演变状态。 本文由滚动轴承性能数据采集开始,依次对性能进行混沌识别,预测未来性能,构建性能可靠性预测模型,由概率密度函数的演变分析性能的特征演变,并对性能可靠性开展评估、假设检验与不确定性分析,形成完整的数据驱动性能可靠性演变预测方法。汽车水泵轴承性能预测的两组案例中,预测可靠度为90%时,对应的轴承性能寿命与实际试验中测量的轴承性能寿命的相对误差分别为3.63%与4.40%,说明性能预测值可信;案例分析中风力发电机主轴轴承第10个时间单元的的双驱统计量差异超过30%,其余时间单元则小于10%,发现了性能可靠性演变的背离点,动力学刻画的可靠性演变过程验证了背离点的正确性。 关键词:滚动轴承;性能;可靠性;演变;预测 论文类型:应用基础研究 选题来源:国家自然科学基金资助项目(项目批准号:51475144),河南省自然科学基金资助项目(项目批准号:162300410065)。
The reliability of rolling bearing during operation has an important impact on the performance of the main engine. Due to the uncertainty of installation conditions, stress wave, lubrication conditions in the process of application, and the special information carried by the bearing itself, the information is not clear, concerning the performance evolution of rolling bearing and failure probability distribution. It is, therefore, difficult to judge the performance reliability with the classical statistical method of the fatigue life probability design based on the stress-strength interference theory. The complexity and changeability of the external environment and internal factors during the operation may lead to the uncertainty of its performance and reliability in the dynamic evolution of the future operation, but the dynamic performance parameters for a specific time series of measured bearing can provide the evolution information of the bearing performance. Special tests are carried out to provide data support for reliability evolution process prediction and test verification. The performance time series are the premise and foundation of research on performance data driven prediction method of rolling bearing reliability evolution process. Considering the rolling bearings of different structure types and the impact of their applications in different stages on performance parameters, working condition simulation bench test and the actual operation performance monitoring test are carried out on the automotive water pump shaft bearing and double row tapered roller bearing on wind power generator spindle, time truncation, fixed performance parameters truncation and rolling bearing performance parameters time series in full life cycle test are obtained. Two kinds of methods of nonlinear dynamic analysis and time series chaotic feature recognition are used to determine the chaotic characteristics of rolling bearing. The typical rolling bearing elements are selected and the nonlinear dynamic equations in the operation of rolling bearing are constructed. With the operating conditions being set, the numerical method are adopted to solve nonlinear equations and to identify the chaotic characteristics of rolling bearing performance. The phase space of the original time series is reconstructed, and the chaotic parameters of calculation performance time series are measured for the verification of chaotic characteristics of bearing performance. The future time series of bearing performance is predicted, and the possible performance life values of bearing are obtained, and the performance reliability prediction model of rolling bearing is established. The comprehensive information of bearing performance evolution cannot be reflected by the forecast results obtained using a single mathematical method, due to the incompleteness of information system or data. Therefore, the performance track evolution characteristics are studied based on the data recorded, and 15 kinds of chaotic forecasting methods are adopted to forecast the trajectory of performance parameters in the future time, and15 new time series are constructed based on the prediction data. The failure value is confirmed and the performance failure domain is delimited according to the needs of working conditions, then performance trend curves corresponding to different prediction methods are obtained. A number of possible bearing performance life values are obtained after the intersection of the failure value and the performance trend curve is got. Grey Bootstrap method is used to carry out Grey Bootstrap for forecast data, the probability density function of rolling bearing performance reliability is established based on histogram constructed with adequate data, for the reliability prediction of data-driven bearing performance. The evolution of the performance reliability of rolling bearing is revealed by the evolution of the probability density function. The probability density function of the rolling bearing performance is established under the condition of maximum entropy based on the maximum entropy principle. Lagrange multiplier is introduced as the characteristic parameters of the probability density function; the original time series are divided into multiple sub time sequences to obtain the probability density function of the rolling bearing performance with different Lagrange multiplier information using life prediction data of different sequences, which can reveal the evolution performance of rolling bearing reliability. Different side information of the probability density function in the evolution process is obtained using hierarchical mining method and subsection mining method. In order to extract the consistent information of the characteristic parameters of the probability density function, the probability density function of the performance reliability of rolling bearing with multiple side information is obtained by the information fusion method. For each order of Lagrange multipliers, Grey Bootstrap is carried out using Lagrange multiplier eigenvalue sets obtained by hierarchical mining and segmentation mining, to obtain the estimated truth value and the upper and lower limits of the confidence interval. The upper and lower limits of the Lagrange multiplier set are combined to obtain the upper and lower limits of performance reliability probability density function, for the evaluation of rolling bearing performance reliability. Hypothesis testing performance of rolling bearing reliability is carried out, and the rolling bearing reliability deviation is explored using hypothesis test double drive statistics, the uncertainty probability model performance of reliability evolution is established to reveal the uncertainty mechanism of reliability evolution. By using the concept of fuzzy equivalent relation, Equivalence relation Probability of rolling bearing performance reliability is obtained. A posteriori probability is established using Bayes statistics to obtain the hypothesis test threshold function to establish the hypothesis test negative domain of the reliability evolution. The double drive statistics is formed using two hypothesis testing methods, and the variation point of the bearing performance reliability is obtained based on double drive statistics quantitative indicators. The uncertain evolution characteristics of reliability true value fluctuation range and its trend, the probability density function of rolling bearing performance to explore the uncertainty mechanism of the performance evolution of rolling bearing. Using kinetic theory, the evolution process of rolling bearing performance reliability is studied respectively from two different angles of velocity and acceleration for the determination of the random evolution state of the uncertainty. This thesis starts with the data acquisition and chaotic identification of performance parameters, followed by the construction of the basic prediction model by forecasting the future performance parameters. The evolution characteristics of the performance parameters is analyzed on the basis of the evolution of the probability density function, and the reliability assessment, hypothesis test and uncertainty analysis are carried out to form a complete forecast method of performance-data-driven reliability evolution. The prediction reliability is 90%, in two groups of cases forecasting the bearing performance of automobile water pump, with 3.63% and 4.40% as the relative error of the corresponding bearing performance life and the bearing life measured in actual tests, which proves the reliability of performance prediction value. The double drive statistic difference is over 30%, concerning the tenth time unit in wind turbine spindle bearing in the case analysis, and the rest of the time unit is less than 10%, thus the departure point of the performance reliability evolution is found and verified by the reliability evolution process by dynamics. KEY WORDS: Rolling bearing; Performance; Reliability; Evolvement; Forecasting Dissertation Type: Applied Fundamental Research Subject Source: National Natural Science Foundation of China (Grant No. 51475144) and Natural Science Foundation of Henan Province of China (Grant No. 162300410065)