基于证据理论的可靠性信息融合方法研究

在对大型客机、导弹、卫星等复杂系统进行可靠性评估时,由于可靠性试验代价昂贵且周期太长,导致可靠性现场试验数据很少,需要充分合理地利用各种来源和各种类型的验前信息,如专家经验、相关型号产品的试验信息及来自不同的试验环境和阶段的信息等,为提高可靠性和置信度的精度,运用Bayes小子样统计推断方法.但Bayes方法应用的一个关键问题就是验前分布的获取和表示问题,提出了利用修正证据组合规则融合多源验前信息的方法,并在融合验前分布的基础上对二项分布产品的可靠性进行了综合评估,并进行仿真,实例表明方法在工程实际中有效果,还可推广应用于其它分布类型的产品的可靠性评估.     

基于混合伽玛分布的Bayes可靠性评估方法

研究指数寿命产品验前信息的融合问题。为了合理地利用验前信息,同时兼顾产品可靠性试验的不确定性,提出运用混合伽玛分布作为验前分布的贝叶斯方法,并将之推广到多源验前信息和变总体的情况。重点解决多源验前信息融合权重的确定问题。通过实例,分析比较这一方法与经典方法和单纯使用伽玛分布作为验前分布的贝叶斯方法的差异,证明该方法的合理性和可行性。     

证据理论在Bayes可靠性评估中的应用

提出了一种基于D-S证据理论的Bayes可靠性评估方案,很好地表达和利用了Bayes方法中的"不确定的"先验信息。在无替换定数截尾试验样本下,研究了指数寿命型产品的可靠性指标的估计问题,提出了可靠性评估的两种方法 :通过融合可靠性指标的后验分布进行综合评估和通过融合可靠性指标的后验置信区间进行综合评估。数值实例表明,利用所提出的方法得到的评估结果是"谨慎的"和合理的。     

基于证据推理和置信规则库的装备寿命评估

针对航天产品试验样本少,寿命评估难的特点,结合产品在研制阶段多种工作环境的失效数据,提出了一种基于证据推理(evidential reasoning,ER)和置信规则库(belief-rule-base,BRB)进行装备寿命评估的新方法.首先,分析了模型的合理性并使用多维BRB模型将多种环境下的寿命数据折合为标准工作环境下的寿命数据,然后通过ER算法将折合后数据和实际工作环境数据进行融合.其次,详细说明了BRB--ER模型的推理过程和寿命评估的步骤.最后,采用某航天产品的失效数据对该方法进行了验证,并用已有的产品寿命的固定值进行BRB的参数更新.研究结果表明,在专家知识准确合理时,该模型能够准确地评估产品寿命,并可根据已有的产品的固定寿命进行训练,建立更加准确的寿命预测模型.     

电子元器件加速退化寿命评估方法研究

基于退化轨迹的评估方法是退化型产品进行可靠性评估的主要方法,适合于对具有退化失效机理的高可靠长寿命产品进行可靠性评估;基于退化轨迹的可靠性评估方法首先选取合适的退化轨迹模型,利用退化数据对退化轨迹进行模型拟合得到模型参数,然后根据退化轨迹外推得到伪失效寿命,最后基于伪失效寿命利用最小二乘法进行统计分析确定产品的失效分布,并通过假设检验的方法选择拟合度最优的分布;本文以大功率开关的加速退化试验数据为例进行了分析和说明.     

增长因子动态模型法Bayes可靠性评定

对于寿命分布为指数分布的可靠性增长试验,将历史数据进行折合,从而运用Bayes方法对失效率进行评估具有重要的意义.研究了增长因子折合法,针对可靠性增长多阶段试验的特点,对采用延缓纠正的可靠性增长模式进行分析,结合AMSAA模型,建立增长因子的动态模型,通过历史数据获得增长因子.运用数据折合和共轭分布,获得当前阶段的验后分布,然后,对多阶段可靠性增长的失效率作Bayes估计,给出了失效率的无偏估计和置信下限.最后通过实例给出了运用该方法进行分析的基本步骤,结果表明该方法具有一定的适用性.     

基于仿真数据的小子样复杂装备系统测试性综合评估

针对小子样复杂装备系统测试性评估问题,提出了一种基于仿真数据的综合评估方法;首先分析了仿真数据的可信性,建立了基于相似度量的仿真数据融合方法;然后,将仿真试验结果作为验前信息,结合实物试验信息确定了测试性的验后分布,并给出了测试性的点估计和置信下限估计;最后通过实例分析表明该方法是合理有效的。     

基于验前试验信息熵的测试性验证试验方案

产品在研制阶段存在大量的试验数据,为有效利用验前数据,降低测试性验证试验样本量,提出一种基于验前试验信息熵的测试性验证试验方案。该方案利用信息熵来度量研制阶段多次验前试验数据对测试性验证试验所起的作用,依据平均互信息熵和信息总量相等的原则,将多次验前试验数据等效成一次成败型数据;在此基础上,通过相容性检验方法确定验前数据与试验数据的相容性水平,并以Beta分布为验前分布,利用加权混合贝叶斯理论建立混合验后分布,之后,基于贝叶斯平均风险理论求解满足双方风险要求的试验方案;最后,以某型雷达发射分机为例,对其进行测试性验证试验研究,研究结果验证了该方案的有效性。     

恒定应力加速寿命试验模型及应用——威布尔分布

加速寿命试验是可靠性试验技术的基本方法之一。检验加速寿命试验中的数据是否服从威布尔分布,进而得出加速系数和加速寿命方程,用在不同应力水平下试验得到的数据,运用威布尔分布模型,并对其中的待估参数进行点估计和区间估计,最终得出产品在正常应力水平下的各项可靠性特征量的估计。     

基于退化数据与marker数据综合的产品可靠性建模分析

基于性能退化数据的可靠性建模分析方法,为现代工业中长寿命、高可靠产品的可靠性研究提供了重要途径,但在破坏性测量中无法获得足够的产品性能退化数据.对此,综合利用产品性能退化数据和marker数据对产品可靠性进行建模,不仅能够提高可靠性评估的精度,还可以在产品运行过程中通过对marker的测量来预测产品剩余寿命,从而为产品维修、更换以及备件决策提供依据.采用二元Wiener过程对产品的性能参数和marker进行建模,给出模型参数的估计方法和基于marker测量数据的剩余寿命预测方法,并通过仿真示例验证所提出方法的有效性.     

Combining expert knowledge and local data for improved service life modeling of water supply networks

The presented approach aims to overcome the scarce data problem in service life modeling of water networks by combining subjective expert knowledge and local replacement data. A procedure to elicit imprecise quantile estimates of survival functions from experts, considering common cognitive biases, was developed and applied. The individual expert priors of the parameters of the service life distribution are obtained by regression over the stated distribution quantiles and aggregated into a single prior distribution. Furthermore, a likelihood function for the commonly encountered censored and truncated pipe replacement data is formulated. The suitability of the suggested Bayesian approach based on elicitation data from eight experts and real network data is demonstrated. Robust parameter estimates could be derived in data situations where frequentist maximum likelihood estimation is unsatisfactory, and to show how the consideration of imprecision and in-between-variance of experts improves posterior inference.     

基于贝叶斯网络的隧道围岩失稳风险预警方法

为了对隧道围岩失稳风险做出准确判断,预防围岩失稳灾害的发生,将贝叶斯网络方法应用于隧道围岩失稳的风险预警。在系统分析隧道支护参数影响围岩稳定性的基础上,通过引入基于贝叶斯网的知识表达和相应的不确定性推理原理,构造了隧道围岩失稳风险预警专家系统;并通过工程案例验证了隧道围岩失稳风险预警专家系统好的适用性。贝叶斯网专家系统可以充分利用专家的先验知识和已有案例提供的或概率分布,可以使推理在输入数据不完备的基础上进行,能够有效地实现围岩失稳风险预警。     

Bayesian dynamic forecasting for attribute reliability

In this paper, we consider a Bayesian dynamic forecasting model that utilizes both the engineering knowledge about the product reliability and attributes (success or failure) data gathered from the inspection of the early stage of development or storage. We assume that a prior distribution of reliability follows a beta distribution. The expected reliability is represented as a cumulative logistic function of the length of time that a new product has been under development or a finished item has been stockpiled in storage. As periodic testing produces attribute data, a prior distribution is updated. The expected reliability and forecasting errors are obtained from a posterior distribution that reflects the uncertainty involved in forecasting. The proposed method is applied to predict the expected reliability decay of a gyroscope in a missile stockpile.     

双不确定相依竞争失效模型的可靠性评估

复杂设备常会经历自然磨损和外力影响,外力致使磨损量突然增大.在工程实践中,很多高可靠性、长寿命的设备,由于不易获得足量的失效数据,须使用专家的经验数据来评估设备的寿命分布和可靠性指标.然而,专家的经验数据往往不是精确的常数,而是一个限定范围.为了处理这种人为认知引起的不确定性,引入不确定理论,用不确定变量刻画专家的经验...     

基于性能退化分析的电连接器可靠性评估

某型电连接器具有寿命长、可靠性高的特点,传统的基于失效寿命数据的可靠性评估方法行不通。提出了基于性能退化分析的可靠性评估方法,给出了进行评估的一般步骤,并以温度为加速应力设计了该型电连接器的加速退化试验。分析了电连接器的接触失效机理,进而推导出退化模型,估计出了样品的伪寿命值。利用Anderson-Darling统计量确定寿命分布类型为对数正态分布,对寿命分布函数和Arrhenius加速方程的参数值进行了极大似然估计,推导出了该型电连接器在工作温度下的可靠度函数,得出t0.9为244 240h。     

Bayesian spatial prediction of the site index in the study of the Missouri Ozark Forest Ecosystem Project

This paper presents a Bayesian spatial method for analysing the site index data from the Missouri Ozark Forest Ecosystem Project (MOFEP). Based on ecological background and availability, we select three variables, the aspect class, the soil depth and the land type association as covariates for analysis. To allow great flexibility of the smoothness of the random field, we choose the Matérn family as the correlation function. We adopt the reference prior as an appropriate prior because there is no previous knowledge of the parameters in the model. An efficient algorithm based on the generalized Ratio-of-Uniforms method is developed for the posterior simulation. One advantage of the algorithm is that it generates independent samples from the required posterior distribution, which is much more efficient for both statistical inference of the parameters and prediction of the site indexes at unsampled locations. Our results show that the aspect class and the soil depth are both significant while the land type association is less significant. The model validation is briefly discussed. In addition, our simulation method allows easy realization for computing quantities from the posterior predictive distributions.     

小数据集的贝叶斯网络结构学习

针对直接基于小数据集贝叶斯网络结构学习不可靠, 以及目前对小数据集的处理只强调扩展而忽略对扩展数据的修正等, 提出了将扩展与修正相结合的小数据集处理机制, 以及在此基础上的基于结点排序和局部打分--搜索的贝叶斯网络结构学习方法. 可不需要完全结点顺序的先验知识, 但能够结合专家的部分结点顺序信息. 实验结果显示了这种方法的有效性和可靠性.     

Restricted exponential forgetting in real-time identification

Identification of time-varying stochastic systems is examined from the Bayesian viewpoint. The questions of which piece of posterior information should be forgotten and of how to forget it are discussed thoroughly, taking into account the main drawbacks of the commonly used technique of exponential forgetting. A novel procedure of suppressing obsolete information is suggested, endowed with two important features. Firstly, it modifies only that piece of so-far accumulated information which is being innovated by the currently measured data. Secondly, it makes it possible to respect partial prior knowledge about the location and/or evolution of parameters in a unified manner. The contributions of the procedure are illustrated in the special case of linear Gaussian regression-type models. A slight, but surprisingly effective, modification of the recursive least-squares method is derived which ensures a high numerical reliability of parameter tracking even for poorly excited systems. Its properties are demonstrated by a simulation example.     

Efficient reliability analysis based on Bayesian framework under input variable and metamodel uncertainties

In the reliability analysis, input variables as well as the metamodel uncertainties are often encountered in practice. The input uncertainty includes the statistical uncertainty of the distribution parameters due to the lack of knowledge or insufficient data. Metamodel uncertainty arises when the response function is approximated by a surrogate function using a finite number of responses to reduce the costly computations. In this study, a reliability analysis procedure is proposed based on a Bayesian framework that can incorporate these uncertainties in an integrated manner into the form of posterior PDF. The PDF, often expressed by arbitrary functions, is evaluated via Markov Chain Monte Carlo (MCMC) method, which is an efficient simulation method to draw random samples that follow the distribution. In order to avoid the nested computation in the full Bayesian approach, a posterior predictive approach is employed, which requires only a single loop of reliability analysis. Gaussian process model is employed for the metamodel. Mathematical and engineering examples are used to demonstrate the proposed method. In the results, comparing with the full Bayesian approach, the predictive approach provides much less information, i.e., only a point estimate of the probability. Nevertheless, the predictive approach adequately accounts for the uncertainties with much less computation, which is more advantageous in the design practice. The smaller the data are provided, the higher the statistical uncertainty, leading to the higher (or lower) failure probability (or reliability).     

用于风险管理的贝叶斯网络学习

结合专家知识和数据进行贝叶斯网络学习.首先利用专家知识建立初始贝叶斯网络结构和参数;然后基于变量之间基本依赖关系、基本结构和依赖分析方法,对初始贝叶斯网络结构进行修正和调整,得到新的贝叶斯网络结构;最后将由专家和数据确定的参数合成为新的参数,得到融合专家知识和数据的贝叶斯网络.该方法可避免现有的贝叶斯网络学习过于依赖数据、对数据的数量和质量要求过高等问题.