EXPLORING REDUCED ORDER MODELS FOR MULTIPHASE FLOWS THAT RETAIN SHARP INTERFACES

Multiphase flows are ubiquitous on Earth. Examples include air interacting with liquid water, sprays in internal combustion engines where liquid fuel is mixed with air, sand dust being carried by blowing wind, and phase change processes such as boiling and cavitation of liquid. It is therefore important to have the capabilities to accurately predict the dynamics of multiphase processes and effectively control them. Direct Numerical Simulations (DNS), where all the length and time scales in the flow field are fully resolved are too computationally costly to deal with the very large systems found in nature and industrial processes, even though computational power has increased immensely in the last couple of decades. Thus, we need coarse grained or reduced order models to speed up the predictions. Large Eddy Simulation (LES) is a promising approach for single phase flow, where the small scales are removed by applying a filter and their effects on the large scales are modeled. Similar ideas can be applied to multiphase flows. For multiphase flows, sharp moving interfaces are often the most distinguished features and in many cases determines their behavior. In a coarse, or reduced order model, it may therefore be important to retain a sharp interface for the resolved scales. Due to a significant progress in machine leaning and data-driven techniques, the closure terms can be modeled by matching the time evolution of the coarse field. This facilitates the development of different reduced order models that are independent of a prior knowledge of the form of closure terms. In this work, we explore different coarse grained representations of complicated multiphase flows that retaining a sharp interface at large scales. We describe a method to singularize a vorticity field, the extraction of a "skeleton" from the interface topology of a jet, as well as a systematic way to coarsen flow fields and interfaces. A data-driven technique to model the closure terms for one of the proposed reduced order models is also presented