Treating Epistemic Uncertainty Using Bootstrapping Selection of Input Distribution Model for Confidence-Based Reliability Assessment

Accurately predicting the reliability of a physical system under aleatory uncertainty requires a very large number of physical output testing. Alternatively, a simulation-based method can be used, but it would involve epistemic uncertainties due to imperfections in input distribution models, simulation models, and surrogate models, as well as a limited number of output testing due to cost. Thus, the estimated output distributions and their corresponding reliabilities would become uncertain. One way to treat epistemic uncertainty is to use a hierarchical Bayesian approach; however, this could result in an overly conservative reliability by integrating possible candidates of input distribution. In this paper, a new confidence-based reliability assessment method that reduces unnecessary conservativeness is developed. The epistemic uncertainty induced by a limited number of input data is treated by approximating an input distribution model using a bootstrap method. Two engineering examples and one mathematical example are used to demonstrate that the proposed method (1) provides less conservative reliability than the hierarchical Bayesian analysis, yet (2) predicts the reliability of a physical system that satisfies the user-specified target confidence level, and (3) shows convergence behavior of reliability estimation as numbers of input and output test data increase.