Multi-source data fusion based small sample prediction of gear random reliability

摘要

In order to predict gear random reliability under the condition of small samples, a model of multi-source data fusion is presented. The gear source data is divided into homologous gear data (HGD) and different source gear data (DSGD) according to their characters. The corresponding algorithms are separately deduced: when in the case of HGD, the grey relational analysis is used to establish the transformation model of gear stress and the model error is considered; when in the case of DSGD, differences in parameters/structure/working conditions are took into account for the purpose of stress transformation. Based on these works, a number of effective stress samples are obtained and distribution parameters of gear stress are estimated by maximum likelihood method. In addition, gear strength reliability is deduced by stress - strength interference model and Monte Carlo sampling. The example shows that gear random reliability can be predicted by work of this study under the condition of small samples; also, accuracy of this method is proved by comparing the result of this work and those of other three methods.