LAUSR

Abstract Model based reliability analysis could be misleading if the simulation model were not validated at the intended design configuration. To improve model accuracy without conceptually revising the model, various model bias correction approaches have been proposed to firstly characterize model bias at training design configurations and then approximate model bias at the intended design configuration. Good accuracy improvement of the model has been shown in literature for not only single model output but also model prediction with multiple or high-dimensional outputs. To date, however, the bias correction approaches are mainly proposed for model prediction with time independent (or static) responses and they cannot be directly applied to the model prediction with time dependent responses. This paper presents such a framework of time dependent model bias correction for model based reliability analysis. In particular, three technical components are proposed including: i) an accuracy metric for time dependent model responses under uncertainty, ii) effective approaches for time dependent model bias calibration and approximation, and iii) reliability analysis considering the time dependent model bias. Two case studies including a structural thermal problem and a corroded beam problem are employed to demonstrate the proposed approach for more effective model based reliability analysis.