A data-driven framework for error estimation and mesh-model optimization in system-level thermal-hydraulic simulation

Using deep learning to explore local physical similarity for global-scale bridging in thermal-hydraulic simulation

“…The predictive capability of FSM for globally extrapolative conditions has been preliminarily evaluated by a turbulent-mixing case study (Bao et al, 2018a) and a two-phase-flow case study (Bao et al, 2019a). A data-driven framework (optimal mesh/model information system, OMIS) was developed and demonstrated to improve applications of the coarse-mesh codes by predicting their simulation errors and suggesting the optimal mesh size and closure models based on the FSM approach (Bao et al, 2019b). This physics-guided datadriven approach is also applied to realize the computationally efficient and scalable CFD prediction of bubbly flow (Bao, Feng et al, 2019).…”

“…Typically, there are two kinds of computational codes used for system thermal-hydraulic analysis: system codes (e.g., RELAP, TRACE) that describe the reactor system as a network of simple control volumes connected with junctions and computational fluid dynamics (CFD)-like codes (e.g., GOTHIC) that provide a three-dimensional (3D) simulation capability using coarse-mesh configurations with the sub-grid phenomena in boundary layer that is well captured by adequate constitutive correlations (e.g., wall functions and turbulence models). Compared with standard system codes (with much loss of local information) and standard fine-mesh CFD codes (with huge computational cost), these coarse-mesh, CFD-like codes have natural advantages and have been widely used to achieve sufficient accuracy for long-term thermal-hydraulic simulation of multicomponent system (Chen et al, 2011, Ozdemir, George & Marshall, 2015, Bao et al, 2016, Bao et al, 2018b. They solve the conservation equations for mass, momentum, and energy for multicomponent multiphase flow.…”

Modelling the effect of computation sampling on insight error in computational fluid dynamics scientific simulation

“…Bao et al learn the relationship between simulation error and the simulation physical features that are derived from the mesh size, model information and simulation parameters. They model this relationship and use the physical features to predict the simulation error to suggest the optimal mesh size for simulation end users given simulation physical features (Bao et al , 2019). Bao et al also use historical simulation data to train a deep forward neural network to predict the simulation error given a grid resolution (Bao et al , 2020).…”

Validation and uncertainty quantification of multiphase-CFD solvers: A data-driven Bayesian framework supported by high-resolution experiments

“…It should be noted that the mesh setup also constitutes a numerical uncertainty source in the MCFD simulation [45]. In this work, a mesh convergence study is performed, and we consider the meshinduced numerical uncertainty is negligible compared to the uncertainty of constitutive relations.…”