Data-Driven Modeling of Rayleigh-Bernard & Vertical Convection in a Two-Phase Cryogenic Fuel Storage Tank Using Sparse Identification of Nonlinear Dynamics

A novel data-driven approach was developed and applied in this study in order to investigate Rayleigh-Bernard and vertical convection in a two-phase cryogenic fuel tank with varying gravity levels. Initially, a pressurized cryogenic tank was simulated using a joint CFD-multinodal technique (CMT), with the gas section developed by a multinode model and the fuel section developed by an axisymmetric CFD solver. As the direct numerical simulation of the liquid part would be computationally expensive, a new data-driven technique called Sparse Identification of Nonlinear Dynamics (SINDy) was utilized to reduce the order of the CFD model. By using the appropriate orthogonal decomposition modes based upon the high-fidelity simulation, SINDy determines the reduced-order model of governing equations. We will link the determined nonlinear compact ordinary differential equations to the multinodal ODE equations for the gas section. We will demonstrate how the SINDy-based CMT overcomes the liability of pure multinodal approaches for capturing the complexity of cryogenic flow dynamics in a two-phase fuel tank under various gravity levels for long-duration investigation.