【作者】 李天梅；

【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】 测试性是装备研制和采办中一个非常重要的技术指标。测试性验证与评估是检验和评估由设计和制造所赋予装备的测试性,是装备采办管理和科学决策的基础。在试验费用和试验周期的约束下,如何开展测试性验证试验设计与综合评估,对装备测试性水平给出低风险的验证结论和高置信度的评估结论,是理论和工程实践中亟待解决的问题,因而具有重要的理论和工程应用价值。论文针对开展测试性验证试验与评估存在的费用高、周期长,验证与评估结论风险高,精度、置信度低等问题,以故障检测率(Fault Detection Rate,FDR)和故障隔离率(Fault Isolation Rate,FIR)为具体的验证与评估指标,提出了基于全寿命周期数据的测试性验证试验优化设计与综合评估技术解决方案,系统深入地研究了测试性验证试验设计中的故障样本优化选取方法、故障有效注入方法和测试性综合评估方法,并通过案例应用,验证方法的有效性。论文的主要研究内容与结论包括:1.在深入分析影响测试性验证试验风险与评估精度、置信度的因素基础上,建立了以二项分布为基础的标准抽样方案与FDR/FIR验证结论风险、置信度的关系模型,给出了抽样样本集大小和样本结构对验证结论风险、置信度的影响规律;建立了故障检测/隔离数据量与FDR/FIR评估结论精度、置信度的关系模型,给出了故障检测/隔离数据量对评估结论精度、置信度的影响规律。上述模型以及所描述的本质关系和规律为后续有针对性的开展测试性验证试验优化设计与综合评估方法研究提供了技术指导并奠定了理论基础。2.为解决由于标准抽样方案确定的故障样本量太大导致试验无法展开;故障样本结构不合理导致测试性验证结论置信度低以及在相同的故障样本量下,由于随机抽样导致故障样本集对被测对象(Unit Under Test,UUT)故障模式集的代表性无法评判等问题,分别从确定优化故障样本量、构建合理的故障样本结构和故障样本集评估优选三个方面,研究并提出了测试性验证试验设计中的故障样本优化选取方法,为装备测试性验证试验设计提供了新的故障样本优化选取思路和方法。(1)针对如何确定优化抽样方案问题,在有效分析研制阶段测试性先验数据基础上,提出了一种利用研制阶段先验数据的优化抽样方案确定方法。研究表明优化抽样方案在承制方、使用方风险基本不变的情况下,能有效减少故障样本量;或在故障样本量保持不变的情况下,将有效降低承制方、使用方风险。(2)针对先验故障率数据不准确导致现有的基于故障率的分层故障样本量分配带来的随机抽样误差过大问题,在分析基于故障率和装备复杂度分层故障样本量分配模型随机抽样误差影响因素基础上,提出了基于专家经验数据的故障率Gamma估计方法和基于定时截尾“小子样”试验数据的故障率Bootstrap极大似然估计方法,据此故障率估计值完成分层故障样本量分配。研究表明本文给出的故障样本量分配得到的故障样本结构更合理,可有效减小随机抽样误差。(3)针对如何对传播型故障模式描述与抽取问题,引入模糊概率Petri网来描述故障的传播扩散过程,并在模糊概率Petri网推理算法基础上求得故障扩散强度。以此为基础,提出了基于故障扩散强度的故障模式随机重要抽样算法。研究表明对故障扩散强度高的故障模式做重要抽样,将能有效降低使用方风险。(4)为解决在相同故障样本量下如何对随机抽取的多个故障样本集进行评估优选,建立了衡量故障样本集对故障模式集代表性好坏的评价指标体系及其量化方法,给出了具体的故障样本集优选模型和方法。研究表明采用本文的优选算法,其优选后的故障样本集能更好地代表被测对象的故障模式集。3.为解决位置不可访问故障的有效注入问题,在深入分析影响故障注入有效性主要因素基础上,建立了故障—状态、故障—故障之间传递特性分析模型,并以故障传递特性为依据,建立了基于故障传递特性的故障模型和位置不可访问故障注入方法以及故障注入策略优化设计方法。研究表明基于故障传递特性的故障注入方法能在保证较高的故障样本注入率,有效节省试验费用的情况下,较好地解决装备位置不可访问故障的注入问题,具有重要的理论意义和工程指导作用。4.针对小样本情况下FDR/FIR评估结论置信度低的问题,研究并提出了基于Bayes变动统计理论的FDR/FIR综合评估模型和方法。首先以多元Dirichlet分布为先验分布,提出了由可更换单元测试性信息、专家经验信息确定先验分布参数的方法,将先验信息转化为先验分布。在此基础上,融合“小子样、异总体”研制阶段增长试验数据和“小子样”外场使用数据,研究并提出了FDR/FIR的Bayes综合评估模型。同时针对FDR/FIR的Bayes综合评估模型的复杂高维后验积分求解问题,分别给出了解析计算方法和马尔科夫链蒙特卡罗法(Markov Chain Monte Carlo,MCMC)抽样计算方法,并仿真分析了模型的稳健性。研究表明利用该模型进行FDR/FIR综合评估,能在较短的外场使用周期内或小样本数据情况下,给出较高置信度的评估结论,为装备测试性综合评估研究提供了重要的理论依据和方法。5.研制开发了装备测试性验证试验优化设计与综合评估系统,并以某型导弹控制系统为对象开展了演示验证应用。该系统能为装备测试性验证试验优化设计与综合评估提供有力的工具,具有很好的工程应用和推广价值。更多还原

【Abstract】 Testability is a critical technical index of the equipment development and purchase. The testability verification test and evaluation are mainly intend to test and evaluate the equipment testability resulting from the design and manufacturing and are the foundation to the equipment purchase management and scientific decision making. How to implement a study on the optimization for the testability verification test and integrated evaluation method considering the limitation of test cost and period and make a low-risk verification conclusion and a high confidence evaluation conclusion is a pending problem in the theory and engineering. It relates to many physical interest and practical applications.This paper is aiming to solve the problems of high cost, long period, high-risk verification/evaluation conclusion and low accuracy/confidence level in the testability verification test and evaluation , which takes the fault detection rate (FDR) and fault isolation rate (FIR) as the specific verification and evaluation index. The systematic scheme for design of testability verification test and integrated evaluation is put forward, which is based on data in whole life period.Then, the failure sampling optimization, failure injection efficiency in the design of testability verification test and testability integrated evaluation methods are studied systematically, and the cases are analyzed to verify the effectiveness of all the methods.The major contents and conclusions of the dissertation are as follows.1. The model of relationship between the plan of sampling and the risk/confidence level of FDR/FIR verification conclusion is established, the relationship between the failure sampling structure and the risk/confidence level is analyzed , the model of relationship between the failure detection/isolation data number and the accuracy/confidence level of FDR/FIR integrated evaluation conclusion is built up by deeply analyzing of the factors which affect the risk, accuracy and confidence level of the testability verification test and evaluation. The models and the natural relationship described can provide guidance for the further pertinent study of testability verification test optimization design and integrated evaluation.2.To settle the problems such as testability verification test unable to perform due to the oversized failure sampling determined by typical plan of sampling, the low confidence level of the testability verification test conclusion resulting from the irrational failure sampling structure and the preferred failure sampling set’s represention for the failure mode set of UUT based on random sampling can not be judged, the failure sample optimization selection method for the design of testability verification test is researched by determining the optimized failure sampling size , building the rational failure sampling structure and failure sampling sets evaluation and optimization. What has been investigated can offer us a new idea and method for the failure sample optimization selection of testability verification test.(1)For the determination of optimized plan of sampling, An optimized plan of sampling which is put forward based on the prior equivalent binomial data during the development stage. The study shows the failure sampling size can be efficiently reduced providing the little risk change of the producer/user and the risk of the producer/user efficiently reduced providing no change of failure sampling size.(2)For the gross error of random sampling of the failure rate-based failure sampling size resulting from the inaccurate prior failure rate, The influencing factors to error of random sampling of the failure rate-based failure sampling size allocation model are analyzed. On such basis, a Gamma distribution-based failure rate estimation method and a Bootstrap maximum likelihood failure rate estimation method are researched, so as to make a failure sampling size allocation distribution on failure rate and equipment complexity basis.The study says a more rational failure sampling structure can be obtained by the failure sampling size allocation on such basis and the error of random sampling can be decreased effectively.(3)For the description and sampling of propagation failure, a fuzzy probability Petri net is built to descript the propagation of failure and the fuzzy probability Petri net inference algorithm is added on the fuzzy probability Petri net to get the failure propagation intensity.On such basis, a random selective sampling algorithm is put forward based on the failure propagation intensity. The study indicates the random selective sampling algorithm can play a key role for the reduction of the final risk of the user.(4)Considering the problems such as evaluate the various failure sampling sets randomly sampled, a evaluation index system is founded and quantified weighing the representation of the failure sampling set to the failure mode set, on such basis the model and method for failure sample set evaluation are put forward. The study indicates the preferred failure sampling set can accurately represent the failure mode set of the subject to be tested.3.For the incapability of effective injection of failure resulting from the inaccessible location, by deepening the analysis on the major factors affecting the effectiveness of the failure injection, the failure propagation characteristic-based failure model and such model-based injection method of failure resulting from the inaccessible location are researched in light of the failure propagation characteristic. The study clarifies the failure propagation characteristic-based failure injection method effectively works out the injection incapability of failure resulting from the inaccessible location. At the same time, the fault injection cost can be considerably cut down. What has been researched can relate to many theory interest and practical applications.4.For problems in the FDR/FIR evaluation occurred when the field usage data for testability of equipment with high reliability is small sample, the Bayes Inference on Dynamic Population-based FDR/FIR integrated evaluation model and method are considered and launched. To comprehensively utilize the plentiful module test information and expert information in the equipment development stage, the pertinent information fusion method is advanced according to various information types. With the multivariate Dirichlet distribution as the prior distribution and the complete fusion of growth test data in development stage and field usage data, the FDR/FIR integrated evaluation model based Bayes inference on dynamic population is set up. An analytical method and Markov Chain Monte Carlo method are adopted to solve the complex posterior distribution inference problem. The model robustness is investigated. The study shows the FDR/FIR integrated evaluation by means of such model can produce the evaluation conclusion with high confidence level in a short field usage period or under“small sample”condition. What has been investigated can offer us an important theory base and method for testability integrated evaluation of equipments.5.A system for optimization of testability verification test and integrated evaluation is developed and take a missile control system as the subject of application. The system furnishes a useful tool for the optimization of testability verification test and integrated evaluation of large scale and complicated equipment.更多还原