A curated dataset for data-driven turbulence modelling

Abstract

The recent surge in machine learning augmented turbulence modelling is a promising approach for addressing the limitations of Reynolds-averaged Navier-Stokes (RANS) models. This work presents the development of the first open-source dataset, curated and structured for immediate use in machine learning augmented corrective turbulence closure modelling. The dataset features a variety of RANS simulations with matching direct numerical simulation (DNS) and large-eddy simulation (LES) data. Four turbulence models are selected to form the initial dataset: k-ε, k-ε-ϕt-f, k-ω, and k-ω SST. The dataset consists of 29 cases per turbulence model, for several parametrically sweeping reference DNS/LES cases: periodic hills, square duct, parametric bumps, converging-diverging channel, and a curved backward-facing step. At each of the 895,640 points, various RANS features with DNS/LES labels are available. The feature set includes quantities used in current state-of-the-art models, and additional fields which enable the generation of new feature sets. The dataset reduces effort required to train, test, and benchmark new corrective RANS models. The dataset is available at 10.34740/kaggle/dsv/2637500.

Super-resolution and uncertainty estimation from sparse sensors of dynamical physical systems

“…An estimate of the epistemic uncertainty of the model was computed by returning two times the standard deviation of the ensemble of predictions at each location. The uncertainty estimate of the model was successful at bounding the error of the prediction mean (EB = 82%) in the low-shot learning environment for the multiple geometry mean flow cases from the McConkey et al (2021) dataset, as listed in Table 1. When applied to the SST and the SM datasets, the percentage of points in the test set where the error was bounded by the uncertainty dropped to 76 and 67%, respectively.…”

“…In practice, due to sensor cost and limitations on the resolution of obtained measurements, data collection is often necessarily constrained, resulting in sparse spatiotemporal measurements with emphasis given to only a few relevant variables. For instance, one of the numerical experiments adopted in this study involves mean fluid flow over multiple surface geometries simulated by time-averaging of Direct Numerical Simulation (DNS)/Large Eddy Simulation (LES) results (McConkey et al, 2021). In order to effectively constrain the learning process with relevant governing equations such as the incompressible Navier-Stokes equations or the Reynolds-averaged Navier-Stokes equations and avoid well-known gradient flow pathologies (Wang S. et al, 2021), the neural network model would need well-resolved simulation or experimental data for each component of the system, namely velocity, pressure, and density.…”

“…Different combinations of network architectures and loss functions were evaluated for different surface geometries from the time-averaged Direct Numerical Simulation (DNS)/RANS dataset of McConkey et al (2021). Table 1 shows the different cases of McConkey et al (2021) used in the present study.…”

“…Different combinations of network architectures and loss functions were evaluated for different surface geometries from the time-averaged Direct Numerical Simulation (DNS)/RANS dataset of McConkey et al (2021). Table 1 shows the different cases of McConkey et al (2021) used in the present study. Three different geometries were chosen: (1) Periodic Hills (PH) that had five realizations of varying steepness but a constant Reynolds number (Re) of 5, 600; (2) a Parametric Bump (PB) with five realizations of different height and a Re between 13, 260 and 27, 850; and (3) a Converging-Diverging channel (CD) with two realizations with a varying Re of 12, 600 and 20, 580.…”

Machine learning and Bayesian inference in nuclear fusion research: an overview

“…Open data sets have been collected in different scientific domains: high-energy physics data at the CERN Open Data portal [125], protein structure database [126], weather forecasting [127,128], and turbulence modeling [129].…”