Calibration of dimensional change in finite element models using AGR moderator brick measurements

Physically based models, resolved using the finite element (FE) method, are often used to model changes in geometry and the associated stress fields of graphite moderator bricks within a reactor. These models require inputs that describe the loading conditions (field variables), and coded relationships describing the behaviour of material properties. Historically, behaviour on material properties have been obtained from Materials Test Reactor (MTR) experiments, however data relating to samples trepanned from operating reactors are increasingly being used to improve models. Geometry measurements from operating reactors offer the potential for improving the coded relationship for dimensional change in FE models. A non-linear mixed-effect model is presented for calibrating the parameters of FE models that are sensitive to mid-brick diameter, using channel geometry measurements obtained from inspection campaigns. The work makes use of a novel technique: the development of a Bayesian emulator, which is a surrogate for the FE model. The use of an emulator allows the influence of the inputs to the finite element model to be evaluated, and delivers a substantial reduction in the computational burden of calibration.