Research On Fuze Reliability Based On Neural Network

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.