Flow, Turbulence and Combustion最新文献

Experiments and numerical modeling on vehicle aerodynamics were conducted in a Reynolds (Re) number one order of magnitude lower than that of typical full-scale application. Drag coefficient, velocity profile measurements and flow visualization were the focus with the proposition of comparing scale effects of a 1:10 sedan passenger vehicle model with numerical data from full-scale (1:1) based on the Reynolds Averaged Navier–Stokes (RANS) approach. After the validation of the numerical approach at 1:10 scale, additional investigation of sharp and rounded fillets presented on the car’s geometry showed to be relevant to the calculation of the separating shear layers and drag prediction, although the general wake structures are qualitatively similar. Effects of the reduced scale are translated to low Reynolds number where viscous effects starts to play a role. Detailed flow features such as recirculating regions and reversing flow acts on the model’s surface while the near wake velocity field is well captured and evaluated both experimentally and numerically. The results permitted to characterize flow details based on Re number flow, to show the effects of sharp corners on the model and to scrutinize the influence of scale effects on vehicle’s aerodynamics.

Most wall shear stress models assume the boundary layer to be fully turbulent, at equilibrium, and attached. Under these strong assumptions, that are often not verified in industrial applications, these models predict an averaged behavior. To address the instantaneous and non-equilibrium phenomenon of separation, the mixture density network (MDN), the neural network implementation of a Gaussian Mixture Model, initially deployed for uncertainty prediction, is employed as a wall shear stress model in the context of wall-modeled large eddy simulations (wmLES) of turbulent separated flows. The MDN is trained to estimate the conditional probability (p(varvec{tau }_wvert textbf{x})), knowing certain entries (textbf{x}), to better predict the instantaneous wall shear stress (varvec{tau }_w) (which is then sampled from the distribution). In this work, an MDN is trained on a turbulent channel flow at the friction Reynolds number (Re_{tau}) of 1000 and on the two-dimensional periodic hill at the bulk Reynolds number of 10,595. The latter test case is known to feature a massive separation from the hill crest. By construction, the model outputs the probability distribution of the two wall-parallel components of the wall shear stress, conditioned by the model inputs: the instantaneous velocity field, the instantaneous and mean pressure gradients, and the wall curvature. Generalizability is ensured by carefully non-dimensionalizing databases with the kinematic viscosity and wall-model height. The relevance of the MDN model is evaluated a posteriori by performing wmLES using the in-house high-order discontinuous Galerkin (DG) flow solver, named Argo-DG, on a turbulent channel flow at (Re_{tau} =2000) and on the same periodic hill flow. The data-driven WSS model significantly improves the prediction of the wall shear stress on both the upper and lower walls of the periodic hill compared to quasi-analytical WSS models.

The extrapolation performance of Convolutional Neural Network (CNN)-based models for Large-Eddy Simulations (LES) has been investigated in the context of turbulent premixed combustion. The study utilises a series of Direct Numerical Simulation (DNS) datasets of turbulent premixed methane/air and hydrogen/air jet flames to train the CNN models. The methane/air flames, which are characterised by increasing Reynolds numbers, are used to model the subgrid-scale flame wrinkling. The hydrogen/air flame, exhibiting complex thermodiffusive instability, is employed to test the ability of the CNN-based combustion models to predict the filtered progress variable source term. This study focuses on the influence of varying training Reynolds numbers, filter sizes, and filter kernels to evaluate the performance of the CNN models to out-of-sample conditions, i.e., not seen during training. The objectives of this study are threefold: (i) analyse the performance of CNN models at different Reynolds numbers compared to the one trained with; (ii) analyse the performance of CNN models at different filter sizes compared to the one trained with; (iii) assess the influence of using different filter kernels (i.e., Gaussian and box filter kernels) between training and testing, to emulate a posteriori applications. The results demonstrate that the CNN models show good extrapolation performance when the training Reynolds number is sufficiently high. Vice versa, when CNN models are trained on low-Reynolds-number flame data, their performance degrades as they are applied to flames with progressively higher Reynolds numbers. When these CNN models are tested on datasets with filter sizes not included in the training process, they exhibit sufficient interpolation capabilities, the extrapolation performance is less precise but still satisfactory overall. This indicates that CNN models can be effectively trained using data filtered with a limited range of filter sizes and then successfully applied across a broader spectrum of filter sizes. Furthermore, when CNNs trained on box-filtered data are applied to Gaussian-filtered data, or vice versa, the models perform well for smaller filter sizes. However, as the filter size increases, the accuracy of the predictions diminishes. Interestingly, increasing the quantity of training data does not significantly enhance model performance. Yet, when training data are distributed with greater weighting towards larger filter sizes, the model’s overall performance improves. This suggests that the strategic selection and weighting of training data can lead to more robust generalization across different filter conditions.

In this study, a duct explosion experiment with an ammonia-methane-air mixture was conducted using a custom-built stainless steel flame acceleration duct (D = 120 mm, L/D = 45.8). The effects of varying ammonia concentrations (φ = 0%, 10%, 20%, 30%) and equivalence ratios (Φ = 0.9, 1.0, 1.1) on flame behavior were examined. The key aspects analyzed included the evolution of the explosion overpressure within the duct and the average propagation velocity of the deflagration flames. The results show that ammonia reduces the brightness of methane-air deflagration flames and that this reduction becomes more pronounced as the ammonia concentration (φ) increases, and the pressure‒time histories inside the duct have a three-peak structure (P_b, P_out, and P_ext), which is caused by the burst of the vent cover, venting of burned mixtures, and counterflow flame generated by the external explosion, Additionally, rarefaction waves in the duct following discharge can lead to oscillatory combustion, and a "backfire" phenomenon is observed in all experiments. This study provides fundamental theoretical support for the promotion and application of ammonia fuel.

The jet-in-coflow is a two-stream configuration having engineering applications in combustors and gas turbine engine exhausts. In practical systems, the coflow generates a boundary layer of the outer wall of the jet pipe and may also have a certain level of turbulence. In the current work, the evolution of this flow configuration is studied using an air-air turbulent jet in a low turbulence coflow (turbulence intensity < 6%), and the 2D velocity field is measured by planar particle image velocimetry. Cases of varying coflow ratio (ratio of coflow velocity to jet velocity) of 0 (turbulent free jet), 0.09, 0.15, and 0.33 are generated by keeping a constant velocity jet (Re = 14000) and varying the coflow velocity. The trends of jet centerline properties such as velocity decay, jet spread, and jet momentum of jet-in-coflow cases, scaled to represent an equivalent free jet, show deviations from that of the turbulent free jet. The radial profile of mean velocity shows a region of velocity deficit, compared to a turbulent free jet, on the coflow side in the jet-in-coflow cases. In contrast, the turbulence intensity and Reynolds shear stress profiles show an enhanced peak near the interface for the jet-in-coflow cases. Further, conditional statistics were extracted by detecting the interface between the jet and the surroundings, wherein the same trends are observed. The low turbulence levels of the coflow have little effect on the jet/coflow interface, as seen by the conditional enstrophy diffusion and tortuosity compared to a turbulent free jet. The differences at the jet/coflow interface of a jet-in-coflow with respect to a turbulent free jet are attributed to the boundary layer initially developed by the turbulent coflow over the pipe generating the jet, and these are seen throughout the near-to-intermediate field (0(le)x/D(le)40).

A Bayesian-based approach is developed to learn predictive turbulence-model corrections for unsteady flow simulations. A distinct feature of the present approach is its ability to perform such a learning task using limited data, which is characteristic of realistic configurations where full sampling can be difficult. Relying on the Expectation–Maximization formalism, the learning task is performed in two steps that optimally combine the strengths of data-assimilation and machine-learning techniques. In a first step, an Ensemble Kalman Filter is used to perform sequential state estimation, namely inferring full flow representations from the considered sparse unsteady data. In a second step, the thus-obtained full states are used to form a training dataset to build the turbulence-model corrections. The present methodology is employed to learn corrective terms for the unsteady Reynolds-Averaged Navier–Stokes (URANS) equations closed by the Spalart–Allmaras model. The sparse data that are used for training are given in the form of a limited number of spatially pointwise velocity observations that are extracted from a Direct Numerical Simulation of the flow past a circular cylinder at (Re=3900). It is shown that the corrected URANS model that is obtained via this strategy significantly outperforms the baseline model despite of the sparse nature of the considered data.

Graphical Abstract

Three complementary approaches for reducing the grid-resolution requirements in hybrid RANS–LES computations, namely (a) the use of wall functions, (b) the application of locally embedded WMLES instead of global WMLES, as well as (c) local grid adaptation in LES regions, are assessed for different test cases up to an industry-relevant aeronautical flow. In this context, targeted improvements and an extension to general 3D geometries of an embedded WMLES method in a second-order accurate, unstructured compressible finite-volume solver are presented. For the wall functions and the embedded WMLES, which are applied to the NASA hump flow and the CRM-HL aircraft configuration, significant computational efficiency gains relative to corresponding reference simulations are demonstrated, while the loss of predictive accuracy compared to experiments can be limited to acceptable levels. Using a refinement indicator based on the locally resolved turbulent kinetic energy, the grid adaptation applied to the NASA hump flow and the NACA0021 at stall conditions yields partly even improved results compared to computations on globally-refined fixed grids, but the computational overhead due to the iterative refinement and averaging process was not yet included in this study. With grid-point savings ranging between 1/3 and more than 2/3 of grid points compared to respective reference meshes, all considered methods offer potential towards more efficient hybrid RANS–LES simulations of complex flows, although their accumulated potential through combination still needs to be explored.

This study presents novel drag reduction active-flow-control (AFC) strategies for a three-dimensional cylinder immersed in a flow at a Reynolds number based on freestream velocity and cylinder diameter of (Re_D=3900). The cylinder in this subcritical flow regime has been extensively studied in the literature and is considered a classic case of turbulent flow arising from a bluff body. The strategies presented are explored through the use of deep reinforcement learning. The cylinder is equipped with 10 independent zero-net-mass-flux jet pairs, distributed on the top and bottom surfaces, which define the AFC setup. The method is based on the coupling between a computational-fluid-dynamics solver and a multi-agent reinforcement-learning (MARL) framework using the proximal-policy-optimization algorithm. This work introduces a multi-stage training approach to expand the exploration space and enhance drag reduction stabilization. By accelerating training through the exploitation of local invariants with MARL, a drag reduction of approximately (9%) is achieved. The cooperative closed-loop strategy developed by the agents is sophisticated, as it utilizes a wide bandwidth of mass-flow-rate frequencies, which classical control methods are unable to match. Notably, the mass cost efficiency is demonstrated to be two orders of magnitude lower than that of classical control methods reported in the literature. These developments represent a significant advancement in active flow control in turbulent regimes, critical for industrial applications.

A joint study utilising experimental and numerical methods was carried out to investigate the aerodynamic effect of a rudder-mounted slat on a vertical stabiliser. The wind tunnel test results showed that the side force coefficient was increased more than 3% with a negligible increase in drag when the rudder deflection angle was set to δ = 30°. Large eddy simulation (LES) results suggested that the rudder-mounted slat can increase the circulation around the vertical stabiliser, showing that the flow from the upstream recirculating regions was drawn towards the rudder surface. Associated changes in the turbulent flow field, including the mean and turbulent flow field and the vortical structure are also presented to understand the flow control mechanism by the rudder-mounted slat.