From DNS to MANN-LES of droplet-laden isotropic turbulence

Abstract

The interaction of liquid droplets with turbulence is relevant to both environmental flows and engineering applications, e.g., rain formation and spray combustion. In this seminar, I will present how we proceeded from studying the physical mechanisms of droplet/turbulence interaction via direct numerical simulation (DNS) to modeling such flow by creating the mixed artificial neural network (MANN) approach for large-eddy simulation (LES).First, in order to perform direct numerical simulation (DNS) of droplet-laden decaying isotropic turbulence, we developed a new pressure-correction method, FastP* (Dodd and Ferrante, 2014), for simulating incompressible two-fluid flows with large density and viscosity ratios between the two phases, coupled with a novel volume-of-fluid (VoF) method (Baraldi et al., 2014). Then, we performed DNS of finite-size droplets of diameter approximately equal to the Taylor length-scale of turbulence in decaying isotropic turbulence (Dodd and Ferrante, 2016). We derived the turbulence kinetic energy (TKE) equations for the two-fluid, carrier-fluid and droplet-fluid flow. This allowed us to explain the pathways for TKE exchange between the carrier turbulent flow and the flow inside the droplet (Dodd and Ferrante, 2016). Next, we developed a new methodology for the spectral analysis of multiphase flows using wavelets (Freund and Ferrante, 2019). We proposed a decomposition of the wavelet energy spectrum into three contributions corresponding to the regions where the wavelet is entirely contained in the carrier phase, entirely contained in a droplet, or partially contained in both carrier and droplet fluids (Freund and Ferrante, 2019). Finally, via analysis of the DNS results both in physical and spectral space, the physical mechanisms we revealed helped us to propose a model for large-eddy simulation (LES) of such flow (Freund and Ferrante, 2021). The main challenge in this endeavor is that the presence of the droplets introduces additional subgrid-scale (SGS) closure terms to the filtered governing equations of motion. The results of a priori analysis showed that they are all significant enough to warrant modeling. Thus, we proposed a new modeling approach that we called mixed artificial neural network (MANN) (Freund and Ferrante, 2021) large-eddy simulation (LES) because it is a mixed LES model that uses the standard Smagorinsky SGS stress model in the carrier flow, and artificial neural networks to predict the SGS closure terms at the interface. Furthermore, we have performed the first a posteriori analysis of such flow for droplets of different Weber numbers showing the agreement of the MANN LES with the filtered DNS results. Finally, the MANN LES approach could be applied to other multiphase turbulent flows due to its ease of implementation, adaptability, and performance.