Uncertainty Quantification (UQ) for the Analysis of Thermal Stresses within a T-junction Pipe Wall Using a Structured Gaussian Process (GP) Method

Abstract

Thermal fatigue of components within the primary and secondary circuits of nuclear power plants (NPPs) is important to understand from a structural integrity perspective. T-junction components within NPPs may be subject to near-wall temperature changes that can induce thermal stresses within pipe walls, eventually leading to fatigue. This paper focusses upon the multiphysics aspects of thermal fatigue phenomena and aims to perform efficient uncertainty quantification (UQ) of one-way coupled, finite volume computational fluid dynamics (CFD) and finite element thermal stress simulations. The aim is to understand the uncertainties associated with the thermal stresses, within the pipe wall of T-junctions, due to uncertainties in transient inlet flow conditions. To minimise the required number of computationally demanding coupled CFD and thermal stress analyses, the structured Gaussian process (GP) based surrogate modelling approach is adopted in this work. The surrogate models of wall temperatures and thermal stresses are trained using the structured GP algorithm with data generated by transient coupled CFD and thermal stress simulations. The GP is then used to propagate the uncertainties in the inlet flow conditions to the temporal evolution of temperatures and stresses at probes located at the pipe inner wall. The GP predictions show that, due to the uncertainties in the transient inlet flow conditions, the maximum temperature and stress can vary between ~210/245°C and ~ −1260/−1660 MPa, respectively, for the temperature transients considered in this study. The proposed GP-based UQ methodology allows for efficient UQ and can be applied to coupled CFD and thermal stress analyses of T-junctions.