Research On Parameters Solution Of Discrete AMSAA Reliability Growing Model Based On EP Algorithm

In the early stages of product development, its reliability and performanceparameters could not immediately reach the desired indicators, it must pass thereliability growth process. The method of measuring the degree of reliability growthis by means of reliability growth model. The reliability growth models widelyaccepted in the reliability field are Duane model and AMSAA model, among whichAMSAA model is divided into continuous and discrete type, widely used for theprocessing and research of various types of data related with the reliability growth.The AMSAA model applies not only to reliability growth testing data tracking, butalso applies to the prediction of reliability growth. The classic method of predictingreliability growth data is drawing reliability growth curve based on AMSAA model,and extrapolating to predict the reliability of the system. Solving reliability growthmodel parameters is the key of reliability growth model.The main solving methods for reliability growth model parameters are LeastSquare method, Figure Forecast method or Maximum Likelihood Estimation method.But there are some common disadvantages to use above methods. Firstly, because ofthe effect of iterative initial value determined by human factors, the above methodsare not easy to converge to the global optimal solution. Secondly, They also needvery more and complicated calculations. The parameter initialization means than theiteration regions are set artificially, so that the results are easily trapped into localminima rather than the global optimum solutions. Also, these algorithms are moredifficult to be applied because of lots of complicated calculations. In this paper,Evolutionary Programming (EP) algorithm is proposed for the model parameterssolution. The process of the model parameters solution based on the EP algorithm notonly avoid the parameter initialization but also no need complicated calculations. The estimation values of parameters will be achieved when the algorithm is executed tosearch the global optimum solutions by its own search process. Also, both theprocess of model parameter solution and the feasibility of the using of EP algorithmin AMSAA model parameter solution is studied by comparison.At last, this thesis studies a reliability growth instance. The above EP algorithmtogether with the experimental data is used to solve the reliability growth modelparameters. The test subjects are470microfarad aluminum electrolytic capacitancesin the computer graphic cards. The experiment consists of two phases: the first stageis that early stale data should be erased by the reliability screening test, and thesecond stage is that the accelerated reliability growth testing is carry out to acquireexperiment data in order to reduced the time required. And finally these data will beusing on the studies of AMSAA model parameter solution based on EP algorithm.