【作者】 赵河明；

【导师】 潘宏侠；

【作者基本信息】 中北大学 ， 火炮、自动武器与弹药工程， 2005， 博士

【摘要】 本论文针对传统的引信可靠性分析与设计中存在的问题,结合国防预研基金项目,进行了基于神经网络的引信可靠性技术研究,研究内容主要如下:1、从引信可靠性分析与设计角度出发,提出了一种应用神经网络技术研究引信可靠性的方法,可用于解决引信在设计和研制中可靠度无法定量化的问题,为我国新型引信的可靠性设计打下基础。2、结合引信的寿命剖面、任务剖面和有关标准,研究了引信可靠性指标的考核方法;依据引信失效事件的同构原理,提出了引信零部件失效形式和状态的两种分类方法,对引信零部件可靠度定量化研究具有重要的意义。3、根据概率论、数理统计和随机过程理论,将传统设计方法中认为是常量的参数作为随机变量来处理,研究了引信几何结构随机变量、材料性能随机变量和环境载荷随机变量的服从分布规律确定方法,随机变量分布规律的确定方法是进行引信可靠性分析和优化设计的基础。4、应用神经网络的函数逼近特性,研究了描述随机变量分布函数和反函数的神经网络方法,并通过网络权值和阈值得到它们的显性表达式,避免了复杂的数值积分运算,并保证分布函数有界性和非减性,为引信可靠性分析与设计提供理论基础。5、依据随机摄动技术和数值逼近法,研究了将引信零件可靠性概率约束转化为确定型约束的可靠性优化设计方法,使其不仅符合引信工作环境要求,而且得出引信可靠性优化设计参数,从而弥补了传统优化设计的不足,使设计方案更加贴近生产实际,是一种具有工程应用价值的综合设计方法。6、对于具有多种失效模式的引信零件,将可靠度简单界限估计法、随机模拟法和神经网络技术相结合,提出了引信零件的可靠性优化设计方法,并以引信用弹簧和传动更多还原

【Abstract】 In this dissertation, fuze reliability technology based on neural networks is studied, whichaims at solving the problem of traditional fuze reliability analysis and design is facing. Thestudy is mainly based on the national defense advanced research fund. The main researchaspects are as follow: First, in the view of the fuze reliability analysis and design, a fuze reliability researchmethod using the neural network is presented, which solves the reliability quantificationproblem of in the process of fuze design and development, expands and develops the fuzereliability theory and lays the foundation for farther fuze reliability research in China. Second, combining the life profile, mission profile with related standard of fuze, theassessing method of fuze reliability index and specifically data being quantified are putforward. According to the isomorphism principle of fuze failure events, two kinds ofclassification method of failure form and state about the components of fuze are put forward,which is of great importance to the reliability quantification research on fuze components. Third, according to probability, mathematical statistics and random process theories, theconstant parameters in traditional design method are processed as the random variables. Thedistribution rule of random variables of geometry,material performance and surroundingsloads are proposed. Therefore, establishing the distribution pattern of random variables is thefoundation of reliability analysis and design optimization of fuze. Fourth, employing the function approximation property of neural networks, a neuralnetwork method is put forward to describe distribution functions and their inverse functions.Then the explicit formula of the distribution functions and their inverse functions are givenusing the network weight value and threshold value, hence preventing from complicatednumerical integral calculation, and ensuring that distribution functions are bounded andmonotone no decreasing functions. At the same time, random variables can be directlysampled conveniently by the neural networks method, which provides theoretical foundationfor fuze reliability analysis and design.Fifth, according to the random perturbation technology and the numerical approximationmethod, an optimum reliability design method is presented, where the probability constraintscan be transformed into deterministic constraint. Not only does this design method meet thework condition of fuze but also the optimum design parameters of fuze reliability isobtained .So this design method perfects traditional optimum design, gets closer to thepractice of production, and is a integrative design method of great value in engineeringapplication.Sixth, for fuze parts with various failure modes, a reliability optimum design method offuze is presented, which combining the reliability estimate theory with a simple boundary,random simulation method with neural network technology. Then the optimum reliabilitydesign process for fuze parts are specified through the examples of spring and drive axle’sdesign.Seventh, the simulation process mathematic model of movement reliability of devicebased on the neural network and Monte Carlo random sampling process are proposed byanalyzing the factors which impact the movement reliability of device. Then, explaining bythe example of the self -adjusting delayed device in fuze, the application of such a method isof great importance for determining reliability of the device and reducing the cost of thedynamic test.Eighth, based on the self organization, self learning and the associated memory of theneural network, the neural network model can be used to predict the storage reliability of thefuze system, which is significant in judging the quality of fuze in stock, predicting thevariation tendency of reliability, and improving the storage reliability of fuze.更多还原