Reliability growth evaluation for highly reliable products based on bayesian dynamic forecasting

For highly reliable products, the last stages of reliability growth tests usually have small sample size, which hampers the corresponding reliability evaluation. A generalized linear model (GLM) is established by the historical multistage samples, and then by Bayes-Monte Carlo method, the estimation of the hyper parameter is obtained to serve as the start point of the growth trend forecasting. In the following test stage, the expectation and variance estimations of the hyper parameter are converted to the prior distribution, and then the hierarchical Bayesian forecasting is carried out to obtain the prior distribution of the failure intensity of the current stage. The finite sample of the current stage is used to perform the Bayesian estimation to obtain the posterior of the hyper parameter, and the evaluation results are calculated recursively. The above Bayesian dynamic forecasting method can effectively make the historical information a supplement of the current stage sample, which is of concise computation form, is applicable to the multistage reliability growth test evaluation of highly reliable products.