A machine-learning based phase change model for simulation of bubble condensation. (October 2021)

• Record Type: Journal Article
• Authors: Tang, Jiguo; Liu, Hongli; Du, Min; Yang, Wei; Sun, Licheng
• Is Part Of: International journal of heat and mass transfer. Volume 178(2021)
• Publication Date: 2021-10
• Subjects: Bubble condensation -- Phase change model -- Volume of fluid -- Artificial neural network; Heat -- Transmission -- Periodicals; Mass transfer -- Periodicals; Chaleur -- Transmission -- Périodiques; Transfert de masse -- Périodiques; Electronic journals; 621.4022
• DOI: 10.1016/j.ijheatmasstransfer.2021.121620
• Languages: English
• ISSNs: 0017-9310

Abstract

Highlights: A new phase change model (ANN-Lee model) for bubble condensation is developed. ANN-Lee model predicts bubble condensation well concern for any empirical factor. ANN-Lee model can be applied to simulate bubble condensation using coarse mesh. The study supports the potential of machine learning for two-phase flow simulation. Abstract: Determining the empirical coefficient and correlation used in phase change model (such as Lee model and empirical correlation model) is one of challenges for the simulation of two-phase flow. To solve this issue in simulation of bubble condensation using volume of fluid (VOF) method, a new phase change model is developed by coupling Lee model and machine learning method, and is termed as ANN-Lee model in this study. The formulation of the empirical coefficient in Lee model is derived based on energy conservation equation. By learning from data collected from previous experiments or generated by empirical correlations, an artificial neural network (ANN) model is trained to calculate the empirical coefficient in each simulation time step. In verified cases of bubble condensation, the ANN-Lee model well predicts the bubble condensation process without concern for the selection of empirical coefficient or correlation in the phase change model. Even using coarse mesh, it can still achieve a comparably accurate prediction as the fine-mesh case. The present simulation results support the feasibility of applying machine learning method for improving the computational fluid dynamics (CFD) prediction of two-phase flow with phase change.