Session R16: CFD: LES, DNS, Hybrid RANS/LES II

Drag Reduction in Flows Past 2D and 3D Circular Cylinders Through Reinforcement Learning

Drag reduction in bluff body flows is crucial for efficient aero- and hydrodynamic design, impacting power consumption and emissions. While passive control methods involve surface modifications, active methods employ surface actuators like tangential actuators, plasma, or mass transpiration. In this study, we identify drag reduction mechanisms in flows past 2D and 3D cylinders controlled by surface actuators using deep reinforcement learning. We investigate flows at Reynolds numbers of 1000, 2000, 4000 (3D), and 8000 (2D). The learning agents are trained in planar flows at various Reynolds numbers, considering actuation energy.

The discovered actuation policies demonstrate remarkable generalization capabilities, enabling open-loop control for Reynolds numbers outside the training range in 2D flows. Surprisingly, the 2D controls and induced drag reduction mechanisms, such as delayed separation and vorticity generation, transfer to three-dimensional cylinder flows. The study unveils intriguing trade-offs between drag reduction and energy input, with aggressive actuation leading to significant decreases in flow separation angle and up to 35\% drag reduction. Conversely, a more conservative approach strategically turns actuators on and off, achieving drag reduction with limited resources.

To alleviate the computational cost associated with stochastic search, we employ a parallel implementation of V-RACER with Remember and Forget for Experience Replay (ReF-ER) in Korali (an open-source optimization and Reinforcement Learning framework), enabling parallel training with direct numerical simulations at a manageable cost. Furthermore, it underscores the cost-effectiveness of surrogate models, such as 2D simulations, for training models applicable to real-world 3D flows at moderately high Reynolds numbers. These findings hold significant implications for drag reduction in practical fluid dynamic applications.   

A Sensitized-RANS, Reynolds-Stress Modeling Study of Flow and Thermal Fields affected by Streamline Curvature

The present study focuses on performing scale-resolving simulations of turbulent flow configurations exhibiting a wide range of differently structured phenomena, including flows influenced by strong streamline curvature, which is characteristic of thermal mixing of flow-crossing streams, and jet impingement on heated walls. These scenarios are commonly encountered in thermotechnical piping systems and internal combustion engines. The anisotropic turbulence residing in the unresolved motion is described by a RANS-based (Reynolds-averaged Navier-Stokes) eddy-resolving closure that accounts for the dynamics of the entire sub-scale stress tensor. The eddy-resolving capability of the model is achieved by introducing an additional production term in the length-scale determining transport equation, whose functional dependence on the second derivative of the underlying velocity field is motivated by the scale-adaptive simulation (SAS) strategy of Menter and Egorov (2010, FTaC 85), Jakirlic and Maduta (2015, IJHFF 51). The flow configurations considered take into account differently structured jets impinging on heated walls, as well as thermal mixing of flow-crossing streams. Comparative evaluation of the results along with available reference experiments and DNS (Direct Numerical Simulation), illustrates the correctly predicted instantaneous character of the flow as well as its time-averaged pattern.   

Regimes of flow over a 6:1 prolate spheroid

The flow around an inclined 6:1 prolate spheroid is a commonly studied canonical problem that exhibits a variety of complex phenomena found in immersed bodies. At angle of attack, the boundary layer separates and rolls up to form a counter-rotating pair of vortices which contribute to a large part of the loads on the spheroid. The present study aims at understanding the effects of Reynolds number and angle of attack regime of separation and the topology of these vortices using large-eddy simulation. A wall-resolved approach is adopted to examine the flow topology for six angles of attack varying from 10 to 90 degrees at Reynolds numbers ranging from 1 million to 4 million based on length and freestream velocity. Different shedding behaviors and vortical structures are observed depending on the two considered parameters, which in turn influence the dynamic loads on the prolate spheroid.   

On the potential of sensitized RANS approaches for unsteady turbulent wall-bounded flows

An improved Scale Adaptive Simulation (SAS) hybrid URANS/LES framework that leverages a sensitized RANS (i.e., k-ε-ξ-f) model to allow for departure-from-equilibrium dynamics is itroduced, primarily to enable the model to transition from URANS to scale resolving mode in attached/mildly separated flows, a known shortcoming of the classical SAS formulations.

Additionally, a transport equation for the length scale of energy-containing eddies is introduced that allows an appropriate transition of length scale from URANS to LES mode. The modified framework is shown to trigger a transition from URANS to LES in stable attached stationary turbulent channel flow and mitigate the log-layer mismatch problem to a great extent. When applied to flows that feature boundary layer separation, i.e., flow over a periodic hill and flow over the wall-mounted hump, the current hybrid framework delivers results in agreement with existing experimental data and comparable to high fidelity LES calculations.   

A Generalized Turbulence Forcing Method to Enable DNS at High Reynolds Numbers

The small-scale behaviour of turbulent flows is often studied using direct numerical simulations (DNS) of homogeneous isotropic turbulence, which need to be forced to achieve high Reynolds numbers and steady state statistics. However, most forcing techniques, meant to represent the effects of scales larger than the DNS domain, forgo the complexity of these energy containing scales and incorporate their effect merely through a single scalar energy injection rate parameter. We introduce a generalized method of forcing turbulence which extracts realistic large scale flow features from large eddy simulations through filtering, and imposes them onto a fully resolved grid in a specific subregion. The method is free from assumptions of equilibrium, isotropy, initial and boundary conditions, and is applicable to any realistic complex flow. The method is evaluated using both a priori and a posteriori analysis, as well as adaptive mesh refinement techniques, and is demonstrated to achieve accuracies comparable to a conventionally forced DNS at a fraction of the cost. Implications of the method for reacting flows involving large upscale transfer of kinetic energy are also discussed.   

Two-Scale Large Eddy Simulation Method for Incompressible Two-Fluid Turbulent Flows

Accurate simulation capabilities for incompressible highly variable density turbulent (IHVDT) flows are critically important for many natural and engineering applications. Developing a large eddy simulation (LES) capability that incorporates all of the relevant subgrid-scale effects is an urgent, currently unmet, scientific and technical challenge. We propose a two-scale framework using the volume-of-fluid method to simulate the incompressible, two-fluid LES problem. This two-scale framework closes the unresolved fluxes in the interface evolution equation through a combination of Lagrangian particles and fractal interpolation techniques, resulting in a mixed Eulerian-Lagrangian methodology.  Our objective is to provide physically accurate surface statistics for the resolved flow, predictions of the unresolved bubble/drop statistics up to the Hinze scale, and to conserve volume for each phase. As a canonical problem, we consider the breakup of a finite-thickness sheet of immiscible fluid immersed in forced homogeneous isotropic turbulence. Using direct numerical simulation, a priori analysis, and a posteriori tests using traditional LES, we point out that neglecting the interface closure terms leads to an under-prediction of the resolved surface statistics during interface breakup as well as the resulting drop/bubble distribution at later stages. We will present analysis and testing of the proposed two-scale LES method to establish the validity and efficacy of the new framework.