Papers
Machine learning for physics, as it is published.
New work from arXiv on CFD surrogates, neural operators and physics-informed learning, refreshed every hour. Search it, filter by topic, or start with the papers we keep coming back to.
Papers
New work from arXiv on CFD surrogates, neural operators and physics-informed learning, refreshed every hour. Search it, filter by topic, or start with the papers we keep coming back to.
564 papers in Turbulence · newest first
physics.flu-dyn
Direct numerical simulation of a turbulent thermal boundary layer (TTBL) can perform the role of an analogy to simulate bushfires that can serve as a testbed for artificial intelligence (AI) enhanced remote sensing of bushfire propagation. By solving the Navier-Stokes equations for a turbulent flow, DNS predicts the flow field and allows for a detailed study of the interactions between the turbulent flow and thermal plumes. In addition to potentially providing insights into the complex bushfire behaviour, direct numerical simulation (DNS) can generate synthetic remote sensing data to train AI algorithms such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), which can process large amounts of remotely sensed data associated with bushfire. Using the results of DNS as training data can improve the accuracy of AI remote sensing in predicting firefront propagation of bushfires. DNS can also test the accuracy of the AI remote sensing algorithms by generating synthetic remote sensing data that allows their performance assessment and uncertainty quantification in predicting the evolution of a bushfire. The combination of DNS and AI can improve our understanding of bushfire dynamics, develop more accurate prediction models, and aid in bushfire management and mitigation.
physics.flu-dyn
Data-driven approaches offer novel opportunities for improving the performance of turbulent flow simulations, which are critical to wide-ranging applications from wind farms and aerodynamic designs to weather and climate forecasting. While conventional continuum Navier-Stokes solvers have been the subject of a significant amount of work in this domain, there has hitherto been very limited effort in the same direction for the more scalable and highly performant lattice Boltzmann method (LBM), even though it has been successfully applied to a variety of turbulent flow simulations using large-eddy simulation (LES) techniques. In this work, we establish a data-driven framework for the LES-based lattice Boltzmann simulation of near-wall turbulent flow fields. We do this by training neural networks using improved delayed detached eddy simulation data. Crucially, this is done in combination with physics-based information that substantially constrains the data-driven predictions. Using data from turbulent channel flow at a friction Reynolds number at , our simulations accurately predict the behaviour of the wall model at arbitrary friction Reynolds numbers up to . In contradistinction with other models that use direct numerical simulation datasets, our physics-informed model requires data from very limited regions within the wall-bounded plane, reducing by three orders of magnitude the quantity of data needed for training. We also demonstrate that our model can handle data configurations when the near-wall grid is sparse. Our physics-informed neural network approach opens up the possibility of employing LBM in combination with highly specific and therefore much more limited quantities of macroscopic data, substantially facilitating the investigation of a wide-range of turbulent flow applications at very large scale.
physics.comp-ph
This paper investigates non-intrusive Scientific Machine Learning (SciML) Reduced-Order Models (ROMs) for plasma turbulence simulations. In particular, we focus on Operator Inference (OpInf) to build low-cost physics-based ROMs from data for such simulations. As a representative example, we consider the (classical) Hasegawa-Wakatani (HW) equations used for modeling two-dimensional electrostatic drift-wave turbulence. For a comprehensive perspective of the potential of OpInf to construct predictive ROMs, we consider three setups for the HW equations by varying a key parameter, namely the adiabaticity coefficient. These setups lead to the formation of complex and nonlinear dynamics, which makes the construction of predictive ROMs of any kind challenging. We generate the training datasets by performing direct numerical simulations of the HW equations and recording the computed state data and outputs the over a time horizon of time units in the turbulent phase. We then use these datasets to construct OpInf ROMs for predictions over additional time units, that is, more than the training horizon. Our results show that the OpInf ROMs capture important statistical features of the turbulent dynamics and generalize beyond the training time horizon while reducing the computational effort of the high-fidelity simulation by up to five orders of magnitude. In the broader context of fusion research, this shows that non-intrusive SciML ROMs have the potential to drastically accelerate numerical studies, which can ultimately enable tasks such as the design of optimized fusion devices.
physics.flu-dyn
Discovering governing equations from data, in particular high dimensional data, is challenging in various fields of science and engineering, and it has potential to revolutionise the science and technology in this big data era. This paper combines sparse identification and deep learning with non-linear fluid dynamics, in particular the turbulent flows, to discover governing equations of nonlinear fluid dynamics in the lower nonlinear manifold space. The autoencoder deep neural network is used to project the high dimensional space into a lower dimensional nonlinear manifold space. The Proper Orthogonal Decomposition (POD) is then used to stabilise the nonlinear manifold space in order to guarantee a stable manifold space for pattern or equations discovery for the highly nonlinear problems such as turbulent flows. Sparse regression is then used to discover the lower dimensional governing equations of fluid dynamics in the lower dimensional nonlinear manifold space. What distinguishes this approach is its ability to discover a lower dimensional governing equations of fluid dynamics in the nonlinear manifold space. We demonstrate this method on a number of high-dimensional fluid dynamic systems such as lock exchange, flow past one and two cylinders. The results demonstrate that the resulting method is capable of discovering lower dimensional governing equations that took researchers in this community many decades years to resolve. In addition, this model discovers dynamics in a lower dimensional manifold space, thus leading to great computational efficiency, model complexity and avoiding overfitting. It also provides a new insight for our understanding of sciences such as turbulent flows.
physics.ao-ph
Optical turbulence presents a significant challenge for communication, directed energy, and imaging systems, especially in the atmospheric boundary layer. Effective modeling of optical turbulence strength is critical for the development and deployment of these systems. The lack of standard evaluation tools, especially long-term data sets, modeling tasks, metrics, and baseline models, prevent effective comparisons between approaches and models. This reduces the ease of reproducing results and contributes to over-fitting on local micro-climates. Performance characterized using evaluation metrics provides some insight into the applicability of a model for predicting the strength of optical turbulence. However, these metrics are not sufficient for understanding the relative quality of a model. We introduce the \texttt{otbench} package, a Python package for rigorous development and evaluation of optical turbulence strength prediction models. The package provides a consistent interface for evaluating optical turbulence models on a variety of benchmark tasks and data sets. The \texttt{otbench} package includes a range of baseline models, including statistical, data-driven, and deep learning models, to provide a sense of relative model quality. \texttt{otbench} also provides support for adding new data sets, tasks, and evaluation metrics. The package is available at \url{https://github.com/cdjellen/otbench}.
physics.flu-dyn
We introduce a novel recursive process to a neural-network-based subgrid-scale (NN-based SGS) model for large eddy simulation (LES) of high Reynolds number turbulent flow. This process is designed to allow an SGS model to be applicable to a hierarchy of different grid sizes without requiring an expensive filtered direct numerical simulation (DNS) data: 1) train an NN-based SGS model with filtered DNS data at a low Reynolds number; 2) apply the trained SGS model to LES at a higher Reynolds number; 3) update this SGS model with training data augmented with filtered LES (fLES) data, accommodating coarser filter size; 4) apply the updated NN to LES at a further higher Reynolds number; 5) go back to 3) until a target (very coarse) filter size divided by the Kolmogorov length scale is reached. We also construct an NN-based SGS model using a dual NN architecture whose outputs are the SGS normal stresses for one NN and the SGS shear stresses for the other NN. The input is composed of the velocity gradient tensor and grid size. Furthermore, for the application of an NN-based SGS model trained with one flow to another flow, we modify the NN by eliminating bias and introducing leaky rectified linear unit function as an activation function. The present recursive SGS model is applied to forced homogeneous isotropic turbulence (FHIT), and successfully predicts FHIT at high Reynolds numbers. The present model trained from FHIT is also applied to decaying homogeneous isotropic turbulence, and shows an excellent prediction performance.
physics.flu-dyn
We investigate the statistical recovery of missing physics and turbulent phenomena in fluid flows using generative machine learning. Here we develop a two-stage super-resolution method using spectral filtering to restore the high-wavenumber components of a Kolmogorov flow. We include a rigorous examination of generated samples through the lens of statistical turbulence. By extending the prior methods to a combined super-resolution and conditional high-wavenumber generation, we demonstrate turbulence recovery on a 8x upsampling task, effectively doubling the range of recovered wavenumbers.
physics.flu-dyn
Turbulence closure modeling using machine learning is at an early crossroads. The extraordinary success of machine learning (ML) in a variety of challenging fields has given rise to justifiable optimism regarding similar transformative advances in the area of turbulence closure modeling. However, by most accounts, the current rate of progress toward accurate and predictive ML-RANS (Reynolds Averaged Navier-Stokes) closure models has been much slower than initially projected. The slower-than-expected rate of progress can be attributed to two reasons: high initial expectations without a complete comprehension of the complexity of the turbulence phenomenon, and the use of ML techniques in a manner that may be inconsistent with turbulence physics. To do full justice to the potential of data-driven approaches in turbulence modeling, this article seeks to identify the foundational physics challenges underlying turbulence closure modeling and assess whether machine learning techniques can effectively tackle them. Drawing analogies with statistical mechanics and stochastic systems, the key physical phenomena and mathematical limitations that render turbulence closure modeling complicated are first identified. The inherent capability of ML to overcome each of the identified challenges is then investigated. The conclusions drawn highlight the limitations of ML-based closures and pave the way for a more judicious framework for the use of ML in turbulence modeling. As ML methods evolve (which is happening at a rapid pace) and our understanding of the turbulence phenomenon improves, the inferences expressed here should be suitably modified.
physics.flu-dyn
Cluster and void formations are key processes in the dynamics of particle-laden turbulence. In this work, we assess the performance of various neural network models for synthesizing preferential concentration fields of particles in turbulence. A database of direct numerical simulations of homogeneous isotropic two-dimensional turbulence with one-way coupled inertial point particles, is used to train the models using vorticity as the input to predict the particle number density fields. We compare autoencoder, U–Net, generative adversarial network (GAN), and diffusion model approaches, and assess the statistical properties of the generated particle number density fields. We find that the GANs are superior in predicting clusters and voids, and therefore result in the best performance. Additionally, we explore a concept of ``supersampling", where neural networks can be used to predict full particle data using only the information of few particles, which yields promising perspectives for reducing the computational cost of expensive DNS computations by avoiding the tracking of millions of particles. We also explore the inverse problem of synthesizing the enstrophy fields using the particle number density distribution as the input at different Stokes numbers. Hence, our study also indicates the potential use of neural networks to predict turbulent flow statistics using experimental measurements of inertial particles.
cs.LG
A data-driven model for predicting the surface temperature using neural networks was proposed to alleviate the computational burden of numerical weather prediction (NWP). Our model, named TPTNet uses only 2m temperature measured at the weather stations of the South Korean Peninsula as input to predict the local temperature at finite forecast hours. The turbulent fluctuation component of the temperature was extracted from the station measurements by separating the climatology component accounting for the yearly and daily variations. The effect of station altitude was then compensated by introducing a potential temperature. The resulting turbulent potential temperature data at irregularly distributed stations were used as input for predicting the turbulent potential temperature at forecast hours through three trained networks based on convolutional neural network (CNN), Swin Transformer, and a graphic neural network (GNN). The prediction performance of our network was compared with that of persistence and NWP, confirming that our model outperformed NWP for up to 12 forecast hours.
physics.flu-dyn
Convolution Neural Networks (CNN) are well-suited to model the nonlinear relationship between the microscale geometry of porous media and the corresponding flow distribution, thereby accurately and efficiently coupling the flow behavior at the micro- and macro- scale levels. In this paper, we have identified the challenges involved in implementing CNNs for macroscale model closure in the turbulent flow regime, particularly in the prediction of the drag force components arising from the microscale level. We report that significant error is incurred in the crucial data preparation step when the Reynolds averaged pressure and velocity distributions are interpolated from unstructured stretched grids used for Large Eddy Simulation (LES) to the structured uniform grids used by the CNN model. We show that the range of the microscale velocity values is 10 times larger than the range of the pressure values. This invalidates the use of the mean squared error loss function to train the CNN model for multivariate prediction. We have developed a CNN model framework that addresses these challenges by proposing a conservative interpolation method and a normalized mean squared error loss function. We simulated a model dataset to train the CNN for turbulent flow prediction in periodic porous media composed of cylindrical solid obstacles with square cross-section by varying the porosity in the range 0.3 to 0.88. We demonstrate that the resulting CNN model predicts the pressure and viscous drag forces with less than 10% mean absolute error when compared to LES while offering a speedup of O(10^6).
physics.flu-dyn
Recent growing interest in using machine learning for turbulence modelling has led to many proposed data-driven turbulence models in the literature. However, most of these models have not been developed with overcoming non-unique mapping (NUM) in mind, which is a significant source of training and prediction error. Only NUM caused by one-dimensional channel flow data has been well studied in the literature, despite most data-driven models having been trained on two-dimensional flow data. The present work aims to be the first detailed investigation on NUM caused by two-dimensional flows. A method for quantifying NUM is proposed and demonstrated on data from a flow over periodic hills, and an impinging jet. The former is a wall-bounded separated flow, and the latter is a shear flow containing stagnation and recirculation. This work confirms that data from two-dimensional flows can cause NUM in data-driven turbulence models with the commonly used invariant inputs. This finding was verified with both cases, which contain different flow phenomena, hence showing that NUM is not limited to specific flow physics. Furthermore, the proposed method revealed that regions containing low strain and rotation or near pure shear cause the majority of NUM in both cases: approximately 76% and 89% in the flow over periodic hills and impinging jet, respectively. These results led to viscosity ratio being selected as a supplementary input variable (SIV), demonstrating that SIVs can reduce NUM caused by data from two-dimensional flows and subsequently improve the accuracy of tensor-basis machine learning models for turbulence modelling.
physics.flu-dyn
We address the problem of data augmentation in a rotating turbulence set-up, a paradigmatic challenge in geophysical applications. The goal is to reconstruct information in two-dimensional (2D) cuts of the three-dimensional flow fields, imagining to have spatial gaps present within each 2D observed slice. We evaluate the effectiveness of different data-driven tools, based on diffusion models (DMs), a state-of-the-art generative machine learning protocol, and generative adversarial networks (GANs), previously considered as the best-performing method both in terms of point-wise reconstruction and the statistical properties of the inferred velocity fields. We focus on two different DMs recently proposed in the specialized literature: (i) RePaint, based on a heuristic strategy to guide an unconditional DM for flow generation by using partial measurements data and (ii) Palette, a conditional DM trained for the reconstruction task with paired measured and missing data. Systematic comparison shows that (i) DMs outperform the GAN in terms of the mean squared error and/or the statistical accuracy; (ii) Palette DM emerges as the most promising tool in terms of both point-wise and statistical metrics. An important property of DMs is their capacity for probabilistic reconstructions, providing a range of predictions based on the same measurements, enabling for uncertainty quantification and risk assessment.
physics.flu-dyn
With the recent rapid developments in machine learning (ML), several attempts have been made to apply ML methods to various fluid dynamics problems. However, the feasibility of ML for predicting turbulence dynamics has not yet been explored in detail. In this study, PredictionNet, a data-driven ML framework based on generative adversarial networks (GANs), was developed to predict two-dimensional (2D) decaying turbulence. The developed prediction model accurately predicted turbulent fields at a finite lead time of up to half the Eulerian integral time scale. In addition to the high accuracy in pointwise metrics, various turbulence statistics, such as the probability density function, spatial correlation function, and enstrophy spectrum, were accurately captured by the employed GAN. Scale decomposition was used to interpret the predictability depending on the spatial scale, and the role of latent variables in the discriminator network was investigated. The good performance of the GAN in predicting small-scale turbulence is attributed to the scale-selection capability of the latent variable. Results also revealed that the recursive applications of the prediction model yielded better predictions than single predictions for large lead times. Furthermore, by utilizing PredictionNet as a surrogate model, a control model named ControlNet was developed to identify disturbance fields that drive the time evolution of the flow field in the direction that optimises the specified objective function. Therefore, an illustrative example in which the evolution of 2D turbulence can be predicted within a finite time horizon and controlled using a GAN-based deep neural network is presented.
physics.flu-dyn
Stratified premixed combustion, known for its capability to expand flammability limits and reduce overall-lean combustion instability, has been widely adopted to comply with increasingly stringent environmental regulations. Numerous numerical simulations with different combustion models and mesh resolutions have been conducted on laboratory-scale flames to further understand the stratified premixed combustion. However, the trade-off between the high-fidelity and low computational cost for simulating laboratory-scale flames still remains, particularly for those combustion models involving direct coupling of chemistry and flow. In the present study, a GPU-based solver is employed to solve partial differential equations and calculate the thermal and transport properties, while an artificial neural network (ANN) is introduced to replace reaction rate calculation. Particular emphasis is placed on evaluating the proposed GPU-ANN approach through the large eddy simulation of the Cambridge stratified flame. The simulation results show good agreement for the flow and flame statistics between the GPU-ANN approach and the conventional CPU-based solver with direct integration (DI). The comparison suggests that the GPU-ANN approach can achieve the same level of accuracy as the conventional CPU-DI solver. In addition, the overall speed-up factor for the GPU-ANN approach is over two orders of magnitude. This study lays the potential groundwork for fully resolved laboratory-scale flame simulations based on detailed chemistry with much more affordable computational cost.
physics.flu-dyn
Turbulent flow over permeable interface is omnipresent featuring complex flow topology. In this work, a data driven, end to end machine learning model has been developed to model the turbulent flow in porous media. For the same, we have derived a non linear reduced order model with a deep convolution autoencoder network. This model can reduce highly resolved spatial dimensions, which is a prerequisite for direct numerical simulation. A downstream recurrent neural network has been trained to capture the temporal trend of reduced modes, thus it is able to provide future evolution of modes. We further evaluate the trained model s capability on a newer dataset with a different porosity. In such cases, fine tuning could reduce the efforts (up to two order of magnitude) to train a model with limited dataset and knowledge and still show a good agreement on the mean velocity profile. Leveraging the current model, we find that even quick fine tuning achieving an impressive order of magnitude reduction in training time by approximately still results in effective flow predictions. This promising discovery encourages the fast development of a substantial amount of data-driven models tailored for various types of porous media. The diminished training time substantially lowers the computational cost when dealing with changing porous topologies, making it feasible to systematically explore interface engineering with different types of porous media. Overall, the data driven model shows a good agreement, especially for the porous media which can aid the DNS and reduce the burden to resolve this complex domain during the simulations. The fine tuning is able to reduce the training cost significantly and maintain an acceptable accuracy when a new flow condition comes into play.
physics.flu-dyn
Despite the increasing availability of high-performance computational resources, Reynolds-Averaged Navier-Stokes (RANS) simulations remain the workhorse for the analysis of turbulent flows in real-world applications. Linear eddy viscosity models (LEVM), the most commonly employed model type, cannot accurately predict complex states of turbulence. This work combines a deep-neural-network-based, nonlinear eddy viscosity model with turbulence realizability constraints as an inductive bias in order to yield improved predictions of the anisotropy tensor. Using visualizations based on the barycentric map, we show that the proposed machine learning method's anisotropy tensor predictions offer a significant improvement over all LEVMs in traditionally challenging cases with surface curvature and flow separation. However, this improved anisotropy tensor does not, in general, yield improved mean-velocity and pressure field predictions in comparison with the best-performing LEVM.
physics.flu-dyn
A multi-fidelity regression model is proposed for combining multiple datasets with different fidelities, particularly abundant low-fidelity data and scarce high-fidelity observations. The model builds upon recent multi-fidelity frameworks based on neural networks, which employ two distinct networks for learning low- and high-fidelity data, and extends them by feeding the gradients information of low-fidelity data into the second network, while the gradients are computed using automatic differentiation with minimal computational overhead. The accuracy of the proposed framework is demonstrated through a variety of benchmark examples, and it is shown that the proposed model performs better than conventional multi-fidelity neural network models that do not use gradient information. Additionally, the proposed model is applied to the challenging case of turbulent flow reconstruction. In particular, we study the effectiveness of the model in reconstructing the instantaneous velocity field of the decaying of homogeneous isotropic turbulence given low-resolution/low-fidelity data as well as small amount of high-resolution/high-fidelity data. The results indicate that the proposed model is able to reconstruct turbulent field and capture small scale structures with good accuracy, making it suitable for more practical applications.
physics.flu-dyn
Turbulent flows have historically presented formidable challenges to predictive computational modeling. Traditional numerical simulations often require vast computational resources, making them infeasible for numerous engineering applications. As an alternative, deep learning-based surrogate models have emerged, offering data-drive solutions. However, these are typically constructed within deterministic settings, leading to shortfall in capturing the innate chaotic and stochastic behaviors of turbulent dynamics. We introduce a novel generative framework grounded in probabilistic diffusion models for versatile generation of spatiotemporal turbulence. Our method unifies both unconditional and conditional sampling strategies within a Bayesian framework, which can accommodate diverse conditioning scenarios, including those with a direct differentiable link between specified conditions and generated unsteady flow outcomes, and scenarios lacking such explicit correlations. A notable feature of our approach is the method proposed for long-span flow sequence generation, which is based on autoregressive gradient-based conditional sampling, eliminating the need for cumbersome retraining processes. We showcase the versatile turbulence generation capability of our framework through a suite of numerical experiments, including: 1) the synthesis of LES simulated instantaneous flow sequences from URANS inputs; 2) holistic generation of inhomogeneous, anisotropic wall-bounded turbulence, whether from given initial conditions, prescribed turbulence statistics, or entirely from scratch; 3) super-resolved generation of high-speed turbulent boundary layer flows from low-resolution data across a range of input resolutions. Collectively, our numerical experiments highlight the merit and transformative potential of the proposed methods, making a significant advance in the field of turbulence generation.
physics.flu-dyn
Deep learning has been employed to identify flow characteristics from Computational Fluid Dynamics (CFD) databases to assist the researcher to better understand the flow field, to optimize the geometry design and to select the correct CFD configuration for corresponding flow characteristics. Convolutional Neural Network (CNN) is one of the most popular algorithms used to extract and identify flow features. However its use, without any additional flow field interpolation, is limited to the simple domain geometry and regular meshes which limits its application to real industrial cases where complex geometry and irregular meshes are usually used. Aiming at the aforementioned problems, we present a Graph Neural Network (GNN) based model with U-Net architecture to identify the vortex in CFD results on unstructured meshes. The graph generation and graph hierarchy construction using algebraic multigrid method from CFD meshes are introduced. A vortex auto-labeling method is proposed to label vortex regions in 2D CFD meshes. We precise our approach by firstly optimizing the input set on CNNs, then benchmarking current GNN kernels against CNN model and evaluating the performances of GNN kernels in terms of classification accuracy, training efficiency and identified vortex morphology. Finally, we demonstrate the adaptability of our approach to unstructured meshes and generality to unseen cases with different turbulence models at different Reynolds numbers.