CONFIDENCE-BASED UNCERTAINTY QUANTIFICATION AND RELIABILITY ASSESSMENT USING LIMITED NUMBERS OF INPUT AND OUTPUT TEST DATA AND VALIDATED SIMULATION MODEL

Accurate reliability assessment requires accurate output distribution. To obtain correct output distribution, a very large number of output physical test data is required, which is prohibitively expensive. Regarding this, simulation-based methods have been developed under the assumption that: (1) accurate input distribution models obtained from large number of input test data; and (2) accurate simulation model (including surrogate model if utilized) that correctly represents physical phenomena. However, in real application, only limited numbers of input test data are available. Thus, input distribution models are uncertain. In addition, the simulation model could be biased due to assumptions and idealizations. Furthermore, only a limited number of physical output test data is available. As a result, a target output distribution to which simulation model can be validated is uncertain and the corresponding reliability is also uncertain. This paper proposes a confidence-based reliability assessment that combines uncertainty due to insufficient input/output test data and biased simulation model. To do that, a hierarchical Bayesian analysis is formulated to obtain uncertainty distribution of reliability. After that, confidence-based reliability is selected at the user-specified target confidence level. It has been numerically demonstrated that the proposed method can estimate reliability of a product that satisfies the user-specified target confidence level.