Towards stochastic modeling for two-phase flow interfacial area predictions: A physics-informed reinforcement learning approach

Abstract

The stochastic nature of turbulent two-phase flow determines that the deterministic modeling approaches always have limited predicting range and accuracy, which is due to the averaging and approximation made during the model developments. On the other hand, well-developed machine learning models that are suitable for complex tasks can be used as a surrogate to improve the predictions. In this paper, a physics-informed reinforcement learning-based method for interfacial area prediction is proposed. The method aims to capture the complexity of the two-phase flow using the advantage of reinforcement learning with the aid of the Interfacial Area Transport Equation. A Markov Decision Process (MDP) that describes the bubble transformation in the fluid flow is established by assuming that the development of two-phase flow is a stochastic process with Markov property. The details of the method’s framework design are described, including the design of the MDP, environment setup, and the algorithms used to solve the RL problem. The performance of the newly proposed method is tested through experiments based on an experimental database for vertical upward bubbly air-water flows. The result shows that with a knowledge-based stochastic policy, good performance of the new method is achieved with the rRMSE of 0.1573, and it is a significant performance boost to the applied IATE model. The approaches to extending the capability of this new RL-based method are discussed, which is a reference for the further development of this approach.