POD–ANN as digital twins for surge line thermal stratification

Abstract

This paper describes a hybrid proper orthogonal decomposition (POD) and artificial neural network (ANN) strategy to construct digital twins of a pressurizer surge line under thermal stratification conditions. The one-way coupled conjugate heat transfer and thermal stress analysis was conducted by use of parametric modeling and the introduction of the inverse distance weighted interpolation for the grid mapping, which allows for the mapped grids to have the same number of nodes regardless of variations of surge line geometries. A snapshot-based POD was utilized to obtain truncated lower-order modes and the full-order system response was projected onto these modes with reduced state coefficients. Then the ANN was employed to establish a surrogate model between the five chosen design variables of interest and the reduced state coefficients, resulting in a surrogate-assisted digital twin for a pressurizer surge line. Prediction of fluid–structure interface temperature and thermal stress distribution was thus achieved in an in-line real-time manner for a wide range of parameter variations. We publicly share all code implementations, and we believe that our efforts open a door for the digital twining of thermo–fluid–structure interaction problems