Direct numerical simulations of multiphase flows: Opportunities and challenges

Abstract

Direct numerical simulations (DNS), where every continuum length and time scale are fully resolved, allow us to follow the evolution of complex flows for sufficiently long time so that meaningful statistical quantities can be gathered. Results for relatively simple multifluid and multiphase systems with bubbles and drops in turbulent flows are now available, but new challenges are emerging. First of all, DNS of very large systems are yielding enormous amount of data that, in addition to providing physical insights, opens up new opportunities for the development of lower order models that describe the average or large-scale behavior. Recent application of machine learning tools to extract closure models from the data for bubbly flows, as well as classify different flow regimes, suggest one possible strategy. Secondly, success with relatively simple systems calls for simulations of more complex problems. Multiphase flows often produce features such as thin films, filaments, and drops that are much smaller than the dominant flow scales and are well-described by analytical or semi-analytical models. Recent efforts to combine semi-analytical models with fully resolved numerical simulations of the rest of the flow include using classical thin film theory for thin films, and boundary layer approximations to compute mass transfer in high Schmidt number bubbly flows.