arXiv subjects

Rahul Agrawal

Publications and source records attributed to Rahul Agrawal.

At least 19 recordsLinked to original sources

arXiv2026

HiLiftAeroML: High-Fidelity Computational Fluid Dynamics Dataset for High-Lift Aircraft Aerodynamics

This paper describes the first-ever open-source high-fidelity CFD dataset of a high-lift aircraft for the purpose of AI surrogate model development. The dataset is composed of 1800 samples, arising from 180 geometry variants and 10 angles of attack for the high-lift NASA Common Research Model (CRM) geometry, used within the AIAA High-Lift Prediction Workshop series. One of the novelties of this dataset is the use of a GPU-accelerated high-fidelity explicit, wall-modeled LES approach for each simulation, using solution-adapted grids between 300M and 500M cells. This ensures the greatest possible accuracy given known challenges in steady-state RANS approaches for these portions of the flight envelope. The entire dataset (geometries, time-averaged volume and surface variables and integral forces) are available, free of charge with a permissive open-source license (CC-BY-4.0). By making this data publicly available, we aim to accelerate the research and development of AI surrogate modeling within the aerospace industry.

physics.flu-dyn↗

arXiv2025

Predictions of flow distortions inside a serpentine diffuser from large-eddy simulations

This work examines the flow separation and the resulting pressure distortions at the exit plane of a serpentine diffuser operating at both subsonic and transonic conditions. Wallmodeled large-eddy simulations (WMLES) using the charLES flow solver are performed at three exit-plane Mach numbers, Ma_AIP ~ {0.36, 0.46, 0.54}. First, it is shown that the onset of flow separation inside a serpentine diffuser may likely experience strong, non-local history effects. A grid refinement study consisting of five grids (from 30 million to 3 billion cells) is conducted for all Mach numbers. The recently proposed dynamic tensor-coefficient Smagorinsky subgrid-scale and sensor-aided non-equilibrium wall models compare favorably with experimental measurements for the pressure recovery and azimuthal flow distortion at all Mach numbers. The pressure recovery and azimuthal flow distortion are predicted to within 0.3% and 7%, respectively, which are both within the experimental error bounds. However, the simulations underpredict the maximum azimuthal distortion in comparison to the experiments for all conditions. Statistical comparisons of the dynamic azimuthal flow distortions suggest that the present LES reasonably captures the ring-averaged mean distortions, and the statistical distributions of the distortion around the mean. Extreme events are underestimated by the present simulations, potentially highlighting that significantly longer integration times may be necessary.

physics.flu-dyn↗

arXiv2025

Direct numerical simulation benchmarks for the prediction of boundary layer bypass transition in the narrow sense

We report a comprehensive set of direct numerical simulation benchmarks of bypass transition in the narrow sense with inlet freestream turbulent intensity levels of 0.75%, 1.5%, 2.25%, 3.0%, and 6.0%, respectively. Detailed descriptions of length scales and the rate of viscous dissipation are provided. We ask two key physical questions. First, how do the decay rates and length scales of freestream turbulence over a transitional and turbulent boundary layer compare to those in spatially developing isotropic turbulence without the wall? Second, what bypass mechanisms drive turbulent spot inception at the intermediate rage of freestream turbulence intensity level? We find that the boundary-layer freestream turbulence decay and length scales evolve similarly to their spatially developing isotropic turbulence flow without the wall counterparts. We also present evidence of the coexistence of two turbulent spot inception mechanisms at the inlet FST level of 2.25%: the long low-speed streak primary and secondary instabilities (only in lower inlet FST levels) and the self-amplifying process of oblique vortex filaments interacting with a Delta-shaped low-speed patch underneath (prevailing only in higher inlet FST levels).

physics.flu-dyn↗

arXiv2025

Development of transitional Reynolds number correlation and assessment of RANS for predictions of bypass transition

We present direct numerical simulations (DNSs) of bypass transition over a flat plate with inlet freestream turbulence intensity levels of 0.75%, 1.5%, 2.25%, 3.0%, and 6.0%, respectively. A new definition of the transition intermittency is proposed based on the mean skin friction. Based on these, we develop an intermittency correlation to predict flow transition. The proposed model is consistent with the classical correlation of Abu-Ghannam and Shaw and reasonably predicts transition Reynolds number (within 10.8% error) for the experiments of Fransson & Shahinfar (2020). Accompanying Reynolds-averaged Navier-Stokes (RANS) simulations for our DNS cases simulations are performed. The RANS results are sensitive to the specification of the inlet turbulence length scale and overpredict (underpredict) the growth of the integral flow scales across the boundary layer during transitional stages when the inlet freestream turbulence is low (high), respectively.

physics.flu-dyn↗

arXiv2025

Large eddy simulation of ocean mesoscale eddies

Mesoscale eddies produce lateral (2D) fluxes that need to be parameterized in eddy-permitting (1/4-degree) global ocean models due to insufficient horizontal resolution. Here, we systematically apply methods from the 3D LES community to parameterize lateral vorticity fluxes produced by mesoscale eddies leveraging an explicit filtering approach together with a dynamic procedure. The developed subfilter closure is implemented into the GFDL MOM6 ocean model and is evaluated in an idealized double-gyre configuration, both a-priori and a-posteriori. For sufficiently resolved grids, the LES simulations converge to the filtered high-resolution data. However, limitations in the proposed closure are observed when the filter scale approaches the energy-containing scales: the a-priori performance drops and a-posteriori experiments fail to converge to the filtered high-resolution data. Nevertheless, the proposed closure is accurate in predicting the mean flow in a-posteriori simulations at all resolutions considered (1/2-1/8 degrees). Finally, we propose parameterizing the thickness fluxes using a Bardina model which further improves simulations at the coarsest resolutions (1/2-1/3 degrees).

physics.ao-ph↗

arXiv2024

Non-equilibrium wall model for large eddy simulations of complex flows exhibiting turbulent smooth body separation

In this work, a non-equilibrium wall model is proposed for the prediction of turbulent flows experiencing adverse pressure gradients, including separated flow regimes. The mean-flow nonequilibrium is identified by comparing two characteristic velocities: the friction velocity (u_tau) and the viscous-pressure gradient velocity (up). In regions where the pressure gradient velocity is comparable to the friction velocity (up \sim u_tau, the near-wall turbulent closure is modified to include the effect of the pressure-gradient and convective terms. The performance of this wall model is evaluated in two canonical flows experiencing smooth body separation: the NASA-Boeing speed bump and the Bachalo-Johnson bump. Improvements in the predictive capabilities of the proposed model for the conventional equilibrium wall model are theorized and then demonstrated through numerical experiments. In particular, the proposed wall model can capture the onset of boundary layer separation observed in experiments or DNS calculations at resolutions where the equilibrium wall model fails to separate.

physics.flu-dyn↗

arXiv2024

Studies of Transonic Aircraft Flows and Prediction of Initial Buffet Onset Using Large-Eddy Simulations

This article utilizes the Large-Eddy Simulation (LES) paradigm with a physics-based turbulence modeling approach, including a dynamic subgrid-scale model and an equilibrium wall model, to examine the flow over the NASA transonic Common Research Model (CRM), a flow configuration that has been the focus of several AIAA Drag PredictionWorkshops (DPWs). The current work explores sensitivities to laminar-to-turbulent transition, wind tunnel mounting system, grid resolution, and grid topology and suggests current best practices in the context of large-eddy simulations of transonic aircraft flows. It is found that promoting the flow transition to turbulence via an array of cylindrical trip dots, including the sting mounting system in the simulations, and leveraging stranded boundary layer grids all tend to improve the quality of the LES solutions. Non-monotonic grid convergence in the LES calculations is observed to be strongly sensitive to grid topology, and stranded meshes rectify this issue relative to their hexagonal close-packed (HCP) counterparts. The details of the boundary layer profiles both at the leading edge of the wing and within the shock-induced separation bubble are studied, with thicknesses and integral measures reported, providing details about the boundary layer characteristics to turbulence modelers not typically available from complex aircraft flows. Finally, an assessment of the initial buffet prediction capabilities of LES is made in the context of a simpler NACA 0012 flow, with computational predictions showing reasonable agreement with available experimental data for the angle of attack at initial buffet onset and shock oscillation frequency associated with sustained buffet.

physics.flu-dyn↗

arXiv2024

SPOT: Point Cloud Based Stereo Visual Place Recognition for Similar and Opposing Viewpoints

Recognizing places from an opposing viewpoint during a return trip is a common experience for human drivers. However, the analogous robotics capability, visual place recognition (VPR) with limited field of view cameras under 180 degree rotations, has proven to be challenging to achieve. To address this problem, this paper presents Same Place Opposing Trajectory (SPOT), a technique for opposing viewpoint VPR that relies exclusively on structure estimated through stereo visual odometry (VO). The method extends recent advances in lidar descriptors and utilizes a novel double (similar and opposing) distance matrix sequence matching method. We evaluate SPOT on a publicly available dataset with 6.7-7.6 km routes driven in similar and opposing directions under various lighting conditions. The proposed algorithm demonstrates remarkable improvement over the state-of-the-art, achieving up to 91.7% recall at 100% precision in opposing viewpoint cases, while requiring less storage than all baselines tested and running faster than all but one. Moreover, the proposed method assumes no a priori knowledge of whether the viewpoint is similar or opposing, and also demonstrates competitive performance in similar viewpoint cases.

cs.RO↗

arXiv2023

Reynolds number dependence of length scales governing turbulent flow separation with application to wall-modeled large-eddy simulations

This article proposes a Reynolds number scaling of the required grid points to perform wall-modeled LES of turbulent flows encountering separation off a solid surface. Based on comparisons between the various time scales in a non-equilibrium (due to the action of an external pressure gradient) turbulent boundary layer, a simple definition of the near-wall ``under-equilibrium" and ``out-of-equilibrium" scales is put forward (where ``under-equilibrium" refers to scales governed by a quasi-balance between the viscous and the pressure gradient terms). It is shown that the former length scale varies with Reynolds number as lp Re^(-2/3). The same scaling is obtained from a simplified Green's function solution of the Poisson equation in the vicinity of the separation point. A-priori analysis demonstrates that the resolution required to reasonably predict the wall-shear stress (for example, errors lower than approximately 10-15% in the entire domain) in several nonequilibrium flows is at least O(10) lp irrespective of the Reynolds number and the Clauser parameter. Further, a series of a-posteriori validation studies are performed to determine the accuracy of this scaling including the flow over the Boeing speed bump, Song-Eaton diffuser, Notre-Dame Ramp, and the backward-facing step. The results suggest that for these flows, scaling the computational grids () such that / lp is independent of the Reynolds number results in accurate predictions of flow separation at the same ``nominal" grid resolution across different Reynolds numbers. Finally, it is suggested that in the vicinity of the separation and reattachment points, the grid-point requirements for wall-modeled large eddy simulations may scale as Re^4/3, which is more restrictive than the previously proposed flat-plate boundary layer-based estimates (Re1) of Choi and Moin (Phys. Fluids, 2012) and Yang and Griffin (Phys. Fluids, 2021).

physics.flu-dyn↗

arXiv2023

An extension of Thwaites method for turbulent boundary layers

Thwaites (1949) developed an approximate method for determining the evolution of laminar boundary layers. The approximation follows from an assumption that the growth of a laminar boundary layer in the presence of pressure gradients could be parameterized solely as a function of a flow parameter, $m = θ^2/ν\frac{dU_e}{ds}$, thus reducing the von Karman momentum integral to a first-order ordinary differential equation. This method is useful for the analysis of laminar flows, and in computational potential flow solvers to account for the viscous effects. However, for turbulent flows, a similar approximation for turbulent boundary layers subjected to pressure gradients does not yet exist. In this work, an approximate method for determining the momentum thickness of a two-dimensional, turbulent boundary layer is proposed. It is shown that the method provides good estimates of the momentum thickness, when compared to available high-fidelity simulation data, for multiple boundary layers including both favorable and adverse pressure gradient effects, up to the point of separation. In the limit of high Reynolds numbers, it is possible to derive a criterion for the onset of separation from the proposed model which is shown to be in agreement with prior empirical observations (Alber, \textit{$9^{th}$ Aerospace Sciences Meeting, 1971}). The sensitivity of the separation location with respect to upstream perturbations is also analyzed through this model for the NASA/Boeing speed bump and the transonic Bachalo-Johnson bump

physics.flu-dyn↗

arXiv2023

Evaluation of eddy viscosity-based models in decaying rotating stratified turbulence

The results of large eddy simulation (LES) using three sub-grid scale models, namely: constant coefficient Smagorinsky, dynamic Smagorinsky, and a dynamic Clark model, for rotating stratified turbulence in the absence of forcing using large-scale isotropic initial condition, are reported here. The LES results are compared to in-house direct numerical simulation (DNS) for establishing grid-independence requirements. Three cases with varying ratios of Brunt-Vaisala frequency to the inertial wave frequency, N / f, have been chosen to evaluate the performance of LES models. The Reynolds number and N / f are chosen as (a) Case 1: Re = 3704, N / f = 5, (b) Case 2: Re = 6667, N / f = 40, and, (c) Case 3: Re = 6667, N / f = 138. This framework is used to illustrate the relative magnitudes of the stratification and rotation which is observed in geophysical flows. Various quantities including turbulent kinetic energy (tke), turbulent potential energy (tpe), total dissipation, potential and total energy spectra, and their fluxes, are analyzed to understand the predictive capability of the various LES models. Results show that all the SGS model predictions are very similar, with the classical Smagorinsky model displaying the highest deviation compared to DNS. The effect of an increase in the value of N / f is also seen in the results of LES with an increase in the oscillations observed in the evolution of tke and tpe and a reduction in dissipation. The spectral analysis shows that the dynamic Clark and Smagorinsky models predict the large-scale physics (κ< 10), while the small scales (10 < κ< 64) energy is under-predicted.

physics.flu-dyn↗

arXiv2023

Stable, entropy-consistent, and localized artificial-diffusivity method for capturing discontinuities

In this work, a localized artificial-viscosity/diffusivity method is proposed for accurately capturing discontinuities in compressible flows. There have been numerous efforts to improve the artificial diffusivity formulation in the last two decades, through appropriate localization of the artificial bulk viscosity for capturing shocks. However, for capturing contact discontinuities, either a density or internal energy variable is used as a detector. An issue with this sensor is that it not only detects contact discontinuities, but also falsely detects the regions of shocks and vortical motions. Using this detector to add artificial mass/thermal diffusivity for capturing contact discontinuities is hence unnecessarily dissipative. To overcome this issue, we propose a sensor similar to the Ducros sensor (for shocks) to detect contact discontinuities, and further localize artificial mass/thermal diffusivity for capturing contact discontinuities. The proposed method contains coefficients that are less sensitive to the choice of the flow problem. This is achieved by improved localization of the artificial diffusivity in the present method. A discretely consistent dissipative flux formulation is presented and is coupled with a robust low-dissipative scheme, which eliminates the need for filtering the solution variables. The proposed method also does not require filtering for the discontinuity detector/sensor functions, which is typically done to smear out the artificial fluid properties and obtain stable solutions. Hence, the challenges associated with extending the filtering procedure for unstructured grids is eliminated, thereby, making the proposed method easily applicable for unstructured grids. Finally, a straightforward extension of the proposed method to two-phase flows is also presented.

physics.flu-dyn↗

arXiv2023

Wall modeled large-eddy simulations of flow over the Sandia Axisymmetric Transonic Bump

Wall-modeled large-eddy simulations (WMLES) are conducted for the flow over the Sandia Axisymmetric Transonic Bump (ATB) at bump chord Reynolds number, Re = 1x10^6 and Mach number, Ma = 0.875. Utilizing various subgrid-scale and wall models, comparisons are made between the simulations and experiments for quantities of engineering interest, such as the skin friction coefficient (Cf) and wall pressure coefficient (Cp) data. Favorable agreements between the simulations and the experiments are achieved, with the simulations predicting the shock location, size, and strength of the separation bubble with reasonable accuracy. In particular, it was established that the realism of the incoming turbulent boundary layer onto the bump is very important for the correct and robust prediction of the skin friction in the upstream region of the bump. Furthermore, a resolution limitation on the predictive capability of the equilibrium wall model in the favorable pressure gradient region is established. Finally, preliminary results utilizing a pressure-gradient inclusive modification to the equilibrium wall model predicts the skin friction within the separation bubble at affordable resolutions

physics.flu-dyn↗

arXiv2022

Non-Boussinesq subgrid-scale model with dynamic tensorial coefficients

A major drawback of Boussinesq-type subgrid-scale stress models used in large-eddy simulations is the inherent assumption of alignment between large-scale strain rates and filtered subgrid-stresses. A priori analyses using direct numerical simulation (DNS) data has shown that this assumption is invalid locally as subgrid-scale stresses are poorly correlated with the large-scale strain rates [Bardina et al., AIAA 1980; Meneveau and Liu, Ann. Rev. Fluid Mech. 2002]. In the present work, a new, non-Boussinesq subgrid-scale model is presented where the model coefficients are computed dynamically. Some previous non-Boussinesq models have observed issues in providing adequate dissipation of turbulent kinetic energy [e.g.: Bardina et al., AIAA 1980; Clark et al. J. Fluid Mech., 1979; Stolz and Adams, Phys. of Fluids, 1999]; however, the present model is shown to provide sufficient dissipation using dynamic coefficients. Modeled subgrid-scale Reynolds stresses satisfy the consistency requirements of the governing equations for LES, vanish in laminar flow and at solid boundaries, and have the correct asymptotic behavior in the near-wall region of a turbulent boundary layer. The new model, referred to as the dynamic tensor-coefficient Smagorinsky model (DTCSM), has been tested in simulations of canonical flows: decaying and forced homogeneous isotropic turbulence (HIT), and wall-modeled turbulent channel flow at high Reynolds numbers; the results show favorable agreement with DNS data. In order to assess the performance of DTCSM in more complex flows, wall-modeled simulations of high Reynolds number flow over a Gaussian bump exhibiting smooth-body flow separation are performed. Predictions of surface pressure and skin friction, compared against DNS and experimental data, show improved accuracy from DTCSM in comparison to the existing static coefficient (Vreman) and dynamic Smagorinsky model.

physics.flu-dyn↗

arXiv2020

Diversifying Relevant Phrases

Diverse keyword suggestions for a given landing page or matching queries to diverse documents is an active research area in online advertising. Modern search engines provide advertisers with products like Dynamic Search Ads and Smart Campaigns where they extract meaningful keywords/phrases from the advertiser's product inventory. These keywords/phrases are representative of a diverse spectrum of advertiser's interests. In this paper, we address the problem of obtaining relevant yet diverse keywords/phrases for any given document. We formulate this as an optimization problem, maximizing the parameterized trade-off between diversity and relevance constrained over number of possible keywords/phrases. We show that this is a combinatorial NP-hard optimization problem. We propose two approaches based on convex relaxations varying in complexity and performance. In the first approach, we show that the optimization problem reduces to an eigen value problem. In the second approach, we show that the optimization problem reduces to minimizing a quadratic form over an l1-ball. Subsequently, we show that this is equivalent to a semi-definite optimization problem. To prove the efficacy of our proposed formulation, we evaluate it on various real-world datasets and compare it to the state-of-the-art heuristic approaches.

cs.IR↗

arXiv2020

XGLUE: A New Benchmark Dataset for Cross-lingual Pre-training, Understanding and Generation

In this paper, we introduce XGLUE, a new benchmark dataset that can be used to train large-scale cross-lingual pre-trained models using multilingual and bilingual corpora and evaluate their performance across a diverse set of cross-lingual tasks. Comparing to GLUE(Wang et al., 2019), which is labeled in English for natural language understanding tasks only, XGLUE has two main advantages: (1) it provides 11 diversified tasks that cover both natural language understanding and generation scenarios; (2) for each task, it provides labeled data in multiple languages. We extend a recent cross-lingual pre-trained model Unicoder(Huang et al., 2019) to cover both understanding and generation tasks, which is evaluated on XGLUE as a strong baseline. We also evaluate the base versions (12-layer) of Multilingual BERT, XLM and XLM-R for comparison.

cs.CL↗

arXiv2019

Turbulent cascade, bottleneck and thermalized spectrum in hyperviscous flows

In many simulations of turbulent flows the viscous forces $ν\nabla^2 {\bf u}$ are replaced by a hyper-viscous term $-ν_p(-\nabla^2)^{p}{\bf u}$ in order to suppress the effect of viscosity at the large scales. In this work we examine the effect of hyper-viscosity on decaying turbulence for values of $p$ ranging from $p=1$ (regular viscosity) up to $p=100$. Our study is based on direct numerical simulations of the Taylor-Green vortex for resolutions from $512^3$ to $2048^3$. Our results demonstrate that the evolution of the total energy $E$ and the energy dissipation $ε$ remain almost unaffected by the order of the hyper-viscosity used. However, as the order of the hyper-viscosity is increased ,the energy spectrum develops a more pronounced bottleneck that contaminates the inertial range. At the largest values of $p$ examined, the spectrum at the bottleneck range has a positive power-law behavior $E(k)\propto k^α$ with the power-law exponent $α$ approaching the value obtained in flows at thermal equilibrium $α=2$. This agrees with the prediction of Frisch et al. [Phys. Rev. Lett. 101, 144501 (2008)] who suggested that at high values of $p$, the flow should behave like the truncated Euler equations (TEE). Nonetheless, despite the thermalization of the spectrum, the flow retains a finite dissipation rate up to the examined order, which disagrees with the predictions of the TEE system implying suppression of energy dissipation. We reconcile the two apparently contradictory results, predicting the value of $p$ for which the hyper-viscous Navier-Stokes goes over to the TEE system and we discuss why thermalization appears at smaller values of $p$.

physics.flu-dyn↗

arXiv2018

Learning From Weights: A Cost-Sensitive Approach For Ad Retrieval

Retrieval models such as CLSM is trained on click-through data which treats each clicked query-document pair as equivalent. While training on click-through data is reasonable, this paper argues that it is sub-optimal because of its noisy and long-tail nature (especially for sponsored search). In this paper, we discuss the impact of incorporating or disregarding the long tail pairs in the training set. Also, we propose a weighing based strategy using which we can learn semantic representations for tail pairs without compromising the quality of retrieval. We conducted our experiments on Bing sponsored search and also on Amazon product recommendation to demonstrate that the methodology is domain agnostic. Online A/B testing on live search engine traffic showed improvements in clicks (11.8\% higher CTR) and as well as improvement in quality (8.2\% lower bounce rate) when compared to the unweighted model. We also conduct the experiment on Amazon Product Recommendation data where we see slight improvements in NDCG Scores calculated by retrieving among co-purchased product.

cs.IR↗