Estimation method for extremely small sample accelerated degradation test data

For the purpose of assessing the accelerated degradation failure time distribution of product, a kind of method to extend extremely small sample for large sample is put forward in this paper. This method divides the original degradation test data into sections and randomly selects partial data from each piece of data. Then, the sampled data form a new test sample. Through this method the sample size could be expended. Considering fixed factors and random factors, the mixed parameter model is adopted to conduct modeling for the degradation paths of the augmented samples. The model parameters are estimated by the two-stage approach. Firstly, each degradation path model parameters are evaluated with least square method. Secondly, the parameters of mixed model are calculated based on the first stage´s estimation. The cumulative probability distribution function of product´s degradation failure time could be derived through the degradation model. The large sample data of natural storage is used to demonstrate this method. The result shows that using piecewise random sampling method to resampling the accelerated test data can effectively expand the test sample size without generating virtual data and solve the problem of evaluating the failure time distribution of degradation test introduced by insufficient samples.