A novel framework for critical heat flux prediction based on multiphase CFD-informed neural networks
Shihao Yang, Xiaoling Li, Bo Ren, Xin Yang, Chong Chen, Qi Lu, Zonglan Wei
Abstract
The Critical Heat Flux (CHF) represents a fundamental thermohydraulic parameter for ensuring the safe operation of thermal systems involving boiling. Recent advancements in multiphase numerical models have established Computational Fluid Dynamics (CFD) as an essential instrument for analyzing boiling phenomena. Nonetheless, the absence of a universal consensus regarding the triggering mechanisms of CHF has resulted in diverse CHF criteria, causing considerable uncertainties in CHF predictions. This study employs a previously developed heat flux partitioning model that directly represents the physical principles of boiling heat transfer across temporal and spatial dimensions. It predicts the CHF for flow boiling in a rectangular channel. A comparative analysis of various CHF criteria on prediction accuracy was conducted. Furthermore, this study applied a neural network framework enhanced with prior domain knowledge derived from the multiphase CFD information. Bayesian optimization was utilized to identify the optimal hyperparameters. A thorough comparative analysis was conducted among the proposed network framework, traditional standalone neural networks, and multiphase CFD methods. This analysis leveraged 433 data points from the 2006 CHF lookup table, encompassing various flow conditions: pressures between 7 and 18 MPa, mass fluxes ranging from 750 to 5000 kg/(m 2 ·s), and qualities varying from -0.2 to 0.25. The results indicate that the multiphase CFD-informed neural network framework exhibits remarkable generalization abilities across the test dataset, particularly regarding extrapolation. Additionally, a comprehensive comparison is presented between the novel framework and the physics-informed machine learning approach, which incorporates empirical correlations as prior knowledge. The proposed multiphase CFD-informed neural network aims to support future experimental efforts and facilitate the analysis of thermodynamic systems pertinent to the boiling crisis.