Reliability estimation of an advanced nuclear fuel using coupled active learning, multifidelity modeling, and subset simulation

Tristructural isotropic (TRISO)-coated particle fuel is a robust nuclear and determining its reliability critical for the success of advanced technologies. However, TRISO failure probabilities are small associated computational models expensive. We used coupled active learning, multifidelity modeling, subset simulation to estimate fuels using several 1D 2D models. With we replaced expensive high-fidelity (HF) model evaluations with information fusion from two low-fidelity (LF) For models, considered three modeling strategies: only Kriging, Kriging LF prediction plus correction, deep neural network (DNN) correction. While results across these strategies compared satisfactorily, employing called HF least often. Next, model, DNN correction (data-driven) (physics-based). The physics-based strategy, as expected, consistently required fewest calls model. data-driven strategy had lower overall time since predictions instantaneous, requires non-negligible time.