An official website of the United States government
The traditional approach of using the Monin–Obukhov similarity theory (MOST) to model near-surface processes in large-eddy simulations (LESs) can lead to significant errors in natural convection. In this study, we propose an alternative approach based on feedforward neural networks (FNNs) trained on output from direct numerical simulation (DNS). To evaluate the performance, we conduct both a priori and a posteriori tests. In the a priori (offline) tests, we compare the statistics of the surface shear stress and heat flux, computed from filtered DNS input variables, to the stress and flux obtained from the filtered DNS. Additionally, we investigate the importance of various input features using the Shapley additive explanations value and the conditional average of the filter grid cells. In the a posteriori (online) tests, we implement the trained models in the System for Atmospheric Modeling (SAM) LES and compare the LES-generated surface shear stress and heat flux with those in the DNS. Our findings reveal that vertical velocity, a traditionally overlooked flow quantity, is one of the most important input features for determining the wall fluxes. Increasing the number of input features improves the a priori test results but does not always improve the model performance in the a posteriori tests because of the differences in input variables between the LES and DNS. Last, we show that physics-aware FNN models trained with logarithmic and scaled parameters can well extrapolate to more intense convection scenarios than in the training dataset, whereas those trained with primitive flow quantities cannot. Significance StatementThe traditional near-surface turbulence model, based on a shear-dominated boundary layer flow, does not represent near-surface turbulence in natural convection. Using a feedforward neural network (FNN), we can construct a more accurate model that better represents the near-surface turbulence in various flows and reveals previously overlooked controlling factors and process interactions. Our study shows that the FNN-generated models outperform the traditional model and highlight the importance of the near-surface vertical velocity. Furthermore, the physics-aware FNN models exhibit the potential to extrapolate to convective flows of various intensities beyond the range of the training dataset, suggesting their broader applicability for more accurate modeling of near-surface turbulence.
more »
« less
An intercomparison of wall fluxes in a turbulent thermal convection chamber: Direct numerical simulations and wall-modeled large-eddy simulations enhanced by machine learning
Thermal convection in a closed chamber is driven by a warm bottom, a cold top, and side walls at various temperatures. Although wall fluxes are the source of convection energy, accurately modeling these fluxes (i.e., the wall model) is challenging. In large-eddy simulations (LESs), many wall models are traditionally derived from the canonical boundary layer, which may be unsuitable for thermal convection bounded by both horizontal and vertical walls. This study conducts a model intercomparison of dry convection in a cubic-meter chamber using three direct numerical simulations (DNSs) and four LESs with different wall models. The LESs employ traditional wall models, a new wall model employing physics-aware neural networks, and a refined grid near the walls. The experiment involves four cases with varying sidewall temperatures. Our results show that LESs capture the main flow features and the trends of mean fluxes. The physics-aware neural networks and refined wall grids can improve the temporally averaged local fluxes when the large-scale circulation has a preferred direction. Even without the local improvement of wall fluxes, the LES flow quantities (temperature and velocities) can still largely match those in DNSs, provided the mean flux largely matches the DNSs. Additionally, DNSs reveal that a variation in corner treatments has minimal impacts on the flow quantities away from corners. Finally, LESs underestimate the mean fluxes of the entire wall due to their inability to resolve corner regions, but their mean flux away from the corner can better match DNS.
more »
« less
- Award ID(s):
- 2133229
- PAR ID:
- 10584223
- Publisher / Repository:
- American Institute of Physics
- Date Published:
- Journal Name:
- Physics of Fluids
- Volume:
- 37
- Issue:
- 4
- ISSN:
- 1070-6631
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
The asymptotic behaviour of Reynolds stresses close to walls is well established in incompressible flows owing to the constraint imposed by the solenoidal nature of the velocity field. For compressible flows, thus, one may expect a different asymptotic behaviour, which has indeed been noted in the literature. However, the transition from incompressible to compressible scaling, as well as the limiting behaviour for the latter, is largely unknown. Thus, we investigate the effects of compressibility on the near-wall, asymptotic behaviour of turbulent fluxes using a large direct numerical simulation (DNS) database of turbulent channel flow at higher than usual wall-normal resolutions. We vary the Mach number at a constant friction Reynolds number to directly assess compressibility effects. We observe that the near-wall asymptotic behaviour for compressible turbulent flow is different from the corresponding incompressible flow even if the mean density variations are taken into account and semi-local scalings are used. For Mach numbers near the incompressible regimes, the near-wall asymptotic behaviour follows the well-known theoretical behaviour. When the Mach number is increased, turbulent fluxes containing wall-normal components show a decrease in the slope owing to increased dilatation effects. We observe that $$R_{vv}$$ approaches its high-Mach-number asymptote at a lower Mach number than that required for the other fluxes. We also introduce a transition distance from the wall at which turbulent fluxes exhibit a change in scaling exponents. Implications for wall models are briefly presented.more » « less
-
Abstract Collisional growth of cloud droplets is an essential yet uncertain process for drizzle and precipitation formation. To improve the quantitative understanding of this key component of cloud‐aerosol‐turbulence interactions, observational studies of collision‐coalescence in a controlled laboratory environment are needed. In an existing convection‐cloud chamber (the Pi Chamber), collisional growth is limited by low liquid water content and short droplet residence times. In this work, we use numerical simulations to explore various configurations of a convection‐cloud chamber that may intensify collision‐coalescence. We employ a large‐eddy simulation (LES) model with a size‐resolved (bin) cloud microphysics scheme to explore how cloud properties and the intensity of collision‐coalescence are affected by the chamber size and aspect ratio, surface roughness, side‐wall wetness, side‐wall temperature arrangement, and aerosol injection rate. Simulations without condensation and evaporation within the domain are first performed to explore the turbulence dynamics and wall fluxes. The LES wall fluxes are used to modify the Scalar Flux‐budget Model, which is then applied to demonstrate the need for non‐uniform side‐wall temperature (two side walls as warm as the bottom and the two others as cold as the top) to maintain high supersaturation in a tall chamber. The results of LES with full cloud microphysics reveal that collision‐coalescence is greatly enhanced by employing a taller chamber with saturated side walls, non‐uniform side‐wall temperature, and rough surfaces. For the conditions explored, although lowering the aerosol injection rate broadens the droplet size distribution, favoring collision‐coalescence, the reduced droplet number concentration decreases the frequency of collisions.more » « less
-
Radiative transfer through clouds can be impacted by variations in particle number size distribution, but also in particle spatial distribution. Due to turbulent mixing and inertial effects, spatial correlations often exist, even on scales reaching the cloud droplet separation distance. The resulting clusters and voids within the droplet field can lead to deviations from exponential extinction. Prior work has numerically investigated these departures from exponential attenuation in absorptive and scattering media; this work takes a step towards determining the feasibility of detecting departures from exponential behavior due to spatial correlation in turbulent clouds generated in a laboratory setting. Large Eddy Simulation (LES) is used to mimic turbulent mixing clouds generated in a laboratory convection cloud chamber. Light propagation through the resulting polydisperse and spatially correlated particle fields is explored via Monte Carlo ray tracing simulations. The key finding is that both mean radiative flux and standard deviation about the mean differ when correlations exist, suggesting that an experiment using a laboratory convection cloud chamber could be designed to investigate non-exponential behavior. Total forward flux is largely unchanged (due to scattering being highly forward-dominant for the size parameters considered), allowing it to be used for conditional sampling based on optical thickness. Direct and diffuse forward flux means are modified by approximately one standard deviation. Standard deviations of diffuse forward and backward fluxes are strongly enhanced, suggesting that fluctuations in the scattered light are a more sensitive metric to consider. The results also suggest the possibility that measurements of radiative transfer could be used to infer the strength and scales of correlations in a turbulent cloud, indicating entrainment and mixing effects.more » « less
-
We introduce a wall model for large-eddy simulation (WMLES) applicable to rough surfaces with Gaussian and non-Gaussian distributions for both the transitionally and fully rough regimes. The model is applicable to arbitrary complex geometries where roughness elements are assumed to be underresolved, i.e. subgrid-scale roughness. The wall model is implemented using a multi-hidden-layer feedforward neural network, with the mean geometric properties of the roughness topology and near-wall flow quantities serving as input. The optimal set of non-dimensional input features is identified using information theory, selecting variables that maximize information about the output while minimizing redundancy among inputs. The model also incorporates a confidence score based on Gaussian process modelling, enabling the detection of potentially low model performance for untrained rough surfaces. The model is trained using a direct numerical simulation (DNS) roughness database comprising approximately 200 cases. The roughness geometries for the database are selected from a large repository through active learning. This approach ensures that the rough surfaces incorporated into the database are the most informative, achieving higher model performance with fewer DNS cases compared with passive learning techniques. The performance of the model is evaluated bothaprioriandaposterioriin WMLES of turbulent channel flows with rough walls. Over 550 channel flow cases are considered, including untrained roughness geometries, roughness Reynolds numbers and grid resolutions for both transitionally and fully rough regimes. Our rough-wall model offers higher accuracy than existing models, generally predicting wall shear stress within an accuracy range of 1%–15 %. The performance of the model is also assessed on a high-pressure turbine blade with two different rough surfaces. We show that the new wall model predicts the skin friction and the mean velocity deficit induced by the rough surface on the blade within 1%–10 % accuracy except the region with transition or shock waves. This work extends the building-block flow wall model (BFWM) introduced by Lozano-Durán & Bae (2023.J. Fluid Mech.963, A35) for smooth walls, expanding the BFWM framework to account for rough-wall scenarios.more » « less
