【作者】 孙锐；

【导师】 黄洪钟；

【作者基本信息】 电子科技大学 ， 机械电子工程， 2012， 博士

【摘要】 随着科学技术的突飞猛进,现代设备和产品朝着大型化、高速化、精密化、一体化的方向快速发展。由于这些设备和产品与人类社会发展及人民生活的联系日益紧密,设备与产品的可靠性越来越受到重视,对可靠性的要求也越来越高。可靠性工程是一个覆盖设备和产品全生命周期的系统工程。可靠性设计、可靠性分配、可靠性评估、可靠性预计、可靠性管理等,都是可靠性工程的范畴。通过几十年的发展,现在已经形成了一整套的可靠性工程方面的理论与方法。传统的可靠性工作是基于可靠性数据的,通过收集的数据,运用以概率论为基础的理论与方法进行分析。但是,在对一些大型、高精密、复杂设备进行可靠性分析时,却面临可靠性数据缺少的问题。造成可靠性数据缺少的原因主要包括设备本身的高可靠以致可以收集到的失效数据较少,设备的高成本以致无法进行寿命实验收集失效数据,设备的结构复杂以致可靠性实验的进行比较困难,前期可靠性工作的缺失以致可靠性数据积累不足。可靠性数据不足称为小样本,小样本无法从概率层面上反映设备的失效规律。在小样本情况下,仍然采用以概率论为基础的可靠性理论与方法进行分析与处理,得到的结果将无法让人信服。其实,在设备和产品的整个生命周期中,除了实验和现场收集的失效数据外,还有很多的数据和信息可以在一定程度上和一定层面上反映设备和产品的可靠性特性。在可靠性数据缺少的情况下,如果能将这些数据和信息进行综合利用,得到的结论将比只利用可靠性数据得到的结论更有说服力。本文以反映设备和产品可靠性特性的这些数据和信息的综合利用为契机,采用以D-S证据理论信息融合的方法对这些数据和信息进行分析,以期得到更能反映设备和产品可靠性水平和特性的结论。本文的主要内容如下:(1)针对现在普遍使用的带权限的D-S证据理论融合算法在证据出现较大冲突时,如果证据权限处于特殊状态,融合结论与常规思维推理结论相悖的问题,通过对证据之间冲突情况的分析,采用集群效应的方法,消弱“坏”对最后融合结论的影响。根据整体冲突程度进行融合权限的再分配。用数值算例对原融合算法和改进后的融合算法的融合结论进行比较分析。(2)D-S证据理论融合算法研究。当辨识框架的维数和证据数量增加的时候,融合计算量将显著增加,特别是在融合过程中加入了权限之后,计算复杂度也会显著增加,这给工程上应用D-S证据理论融合算法进行信息快速融合带来了困难。针对这一问题,通过向量空间分析,将辨识框架的可能假设转化成向量空间中相互垂直的方向,利用向量空间中向量的相加来处理证据融合问题。(3)针对D-S证据理论融合算法应用中,基本概率分布的确定时,采用统计数据构造会忽略掉原始统计数据的分布信息问题,将数据分布信息纳入到融合算法中,通过评价融合结论的分布情况来评定融合结论的可信度,并为多层信息融合提供层次间基本概率分配(Basic Probability Assignments,BPAs)和分布信息传递方法。(4)针对可靠性数据处理中的寿命数据回归分析中原始数据出现明显的分组状况,将分组情况所包含的信息纳入到融合过程中,提出了样本特征预处理方法(Sample Character Preprocess Method,SCPM)。在此基础上提出了可靠性数据和信息融合框架。(5)以某柴油发动机涡轮增压器的FMECA分析中的风险评估问题为例,在可靠性数据缺少的情况下,以专家评判信息为基础,用风险优先数的方法进行风险评价和排序。分别用D-S证据理论、向量空间分析为基础的简便融合方法和可靠性信息融合框架方法进行计算分析,结果表明,在工程实际中,信息融合可以在不同的层次进行,选择不当可能造成融合结论不可用。更多还原

【Abstract】 The development of science and technology lead modern device to large-scale, high-speed, precision and integration. Because the relationship between device and human social development and people life, the reliability of equipment and product gets more and more attention, the requirement to reliability are also getting higher.Reliability engineering is an engineering system through device and produce life cycle. Reliability design, reliability allocation, reliability evaluation, reliability prediction and reliability management are the entire reliability engineering category. With decades of development, people have built a complete reliability theory and method system. Traditional reliability work is based on reliability data, and the theory and method are based on Probability theory.Unfortunately, we face the problem of lack of reliability data with some large scale, high precision, and complex equipment. The reason of lack of reliability data including high reliability leading lack of failure records, high value leading life experiment hard to be carry on, the defect of reliability work leading reliability data accumulation insufficiency.The situation of lack of reliability is called small sample, small sample can not reflect the failure regularity and status on the probability level. If we handle the reliability data with traditional reliability theory and method in the situation of small sample, the analysis result is convincing.In fact, in the whole life cycle of device and product, beside the failure data collected in the experiments and in the using period, there are more data and information which could reflect the device and product’s reliability characters could be got. If we can get these data and information, and make use of these data and information in situation of small sample, we must get more convincing conclusion than the tradition method.This paper proposes to make use of these useful reliability data and information, and get more convincing conclusion based on D-S evidence theory. The main work and contribution of this paper is as follow:(1) Traditional weighted D-S evidence theory combination rule has a unavoidable shortage, when the evidence conflict level is high, and the weight factors of evidence is in a special situation, the fusion result may be contrary to the rational thinking. This paper use cluster effect reduce the influence of“bad”evidence through the analysis to the evidence conflict, reassign weight factor according to the whole conflict level. And compare the fusion result of traditional evidence combination rule and the improved one.(2) Fusion information with D-S evidence theory combination rules, the computation quantity will increase fast when the number of hypothesis and evidence increase, especially in the weighted evidence theory combination rules. This character lead D-S evidence theory combination rule is hard to be implying in engineering practice in some situation. Through vector space analysis, this paper propose a simple information fusion method, which converse the hypothesis in evidence theory into the vertical directions in vector space, and use the vector adding method the handle the information fusion problem.In evidence theory, the distribution information will be ignored in the determination of BPAs with statistics data. To avoid this shortage, this paper proposes a method which brings the distribution information into the fusion process, and evaluates the fusion result’s credibility with the distribution of fusion result. This method also gives the BPAs and distribution information transmission way in multilayer fusion structure.In reliability data analysis practice, we face a situation that the collected data could be grouped in some sets, and there is clear difference among these sets which could not be ignored. This paper proposes a sample character preprocess method, and make use of these sets information in the fusion procedure. And propose the reliability data and information fusion framework on the base of SCPM.In the FMECA analysis to a diesel turbocharger practice, we can not get enough reliability data, and use experts’judgments to risk analysis. With these expert judgments, we use the traditional D-S evidence theory combination rule, simple information fusion method based on vector space analysis and the reliability data and information fusion framework to handle these expert judgments, compare and analysis the results of these three method.更多还原