Computers & Fluids最新文献_第6页

A Python-based flow solver for numerical simulations using an immersed boundary method on single GPUs

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-12-09 DOI: 10.1016/j.compfluid.2024.106511

M. Guerrero-Hurtado , J.M. Catalán , M. Moriche , A. Gonzalo , O. Flores

We present an efficient implementation for the simulation of three-dimensional, incompressible flow around moving bodies with complex geometries on single GPUs, based on Nvidia CUDA through Numba and Python. The flow is solved in this framework through an implementation of the Immersed Boundary Method tailored for the GPU, where different GPU grid architectures are exploited to optimize the overall performance. By targeting a single-GPU, we eliminate the need for both intra- and inter-node communication, which can potentially introduce overheads. With this approach, all simulation data remains in the GPU’s global memory at all times. We provide details about the numerical methodology, the implementation of the algorithm in the GPU and the memory management, critical in single-GPU implementations. Additionally, we verify the results comparing with our analogous CPU-based parallel solver and assess satisfactorily the efficiency of the code in terms of the relative computing time of the different operations and the scaling of the CPU code compared to a single GPU case. Overall, our tests show that the single-GPU code is between 34 to 54 times faster than the CPU solver in peak performance (96–128 CPU cores). This speedup mainly comes from the change in the method of solution of the linear systems of equations, while the speedup in sections of the algorithm that are equivalent in the CPU and GPU implementations is more modest (i.e.,

\times 1.6 - 3

speedup in the computation of the non-linear terms). Finally, we showcase the performance of this new GPU implementation in two applications of interest, one for external flows (i.e., bioinspired aerodynamics) and one for internal flows (i.e., cardiovascular flows), demonstrating the strong scaling of the code in two different GPU cards (hardware).

{"title":"A Python-based flow solver for numerical simulations using an immersed boundary method on single GPUs","authors":"M. Guerrero-Hurtado , J.M. Catalán , M. Moriche , A. Gonzalo , O. Flores","doi":"10.1016/j.compfluid.2024.106511","DOIUrl":"10.1016/j.compfluid.2024.106511","url":null,"abstract":"<div><div>We present an efficient implementation for the simulation of three-dimensional, incompressible flow around moving bodies with complex geometries on single GPUs, based on Nvidia CUDA through Numba and Python. The flow is solved in this framework through an implementation of the Immersed Boundary Method tailored for the GPU, where different GPU grid architectures are exploited to optimize the overall performance. By targeting a single-GPU, we eliminate the need for both intra- and inter-node communication, which can potentially introduce overheads. With this approach, all simulation data remains in the GPU’s global memory at all times. We provide details about the numerical methodology, the implementation of the algorithm in the GPU and the memory management, critical in single-GPU implementations. Additionally, we verify the results comparing with our analogous CPU-based parallel solver and assess satisfactorily the efficiency of the code in terms of the relative computing time of the different operations and the scaling of the CPU code compared to a single GPU case. Overall, our tests show that the single-GPU code is between 34 to 54 times faster than the CPU solver in peak performance (96–128 CPU cores). This speedup mainly comes from the change in the method of solution of the linear systems of equations, while the speedup in sections of the algorithm that are equivalent in the CPU and GPU implementations is more modest (i.e., <span><math><mrow><mo>×</mo><mn>1</mn><mo>.</mo><mn>6</mn><mo>−</mo><mn>3</mn></mrow></math></span> speedup in the computation of the non-linear terms). Finally, we showcase the performance of this new GPU implementation in two applications of interest, one for external flows (i.e., bioinspired aerodynamics) and one for internal flows (i.e., cardiovascular flows), demonstrating the strong scaling of the code in two different GPU cards (hardware).</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106511"},"PeriodicalIF":2.5,"publicationDate":"2024-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138730","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

LES study of turbulent flow fields over a three-dimensional steep hill considering the effects of thermal stratification

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-12-09 DOI: 10.1016/j.compfluid.2024.106521

Tong Zhou, Takeshi Ishihara

In this study, large-eddy simulations are performed to elucidate the spatiotemporal characteristics and physical mechanisms of turbulent boundary layers over hilly terrain under stable, neutral, and unstable stratification. The impact of thermal stratification on turbulent flows over a steep three-dimensional hill is clarified through flow patterns and statistical characteristics. Compared to neutral stratification, the separation bubble downstream of the hill crest is reduced under unstable stratification, while it is enlarged under stable stratification. In addition, turbulent eddy motions in the wake region are enhanced in the unstable condition but are suppressed in the stable condition. Both mean velocities and turbulence fluctuations over steep hilly terrain are amplified by unstable stratification and attenuated by stable stratification. The flow characteristics on the hill crest are comprehensively determined by the topography and thermal stratification, whereas the flow dynamics in the hill wake are predominantly influenced by terrain-induced turbulence. Moreover, the mechanisms driving the formation of flow fields over steep hilly topography under different thermal stratification are investigated through force balance analysis using the time-averaged Navier–Stokes equations. The results indicate that turbulence plays a negligible role in the force balance upstream of the hill, while it becomes the dominant factor for the force balances downstream of the hill.

{"title":"LES study of turbulent flow fields over a three-dimensional steep hill considering the effects of thermal stratification","authors":"Tong Zhou, Takeshi Ishihara","doi":"10.1016/j.compfluid.2024.106521","DOIUrl":"10.1016/j.compfluid.2024.106521","url":null,"abstract":"<div><div>In this study, large-eddy simulations are performed to elucidate the spatiotemporal characteristics and physical mechanisms of turbulent boundary layers over hilly terrain under stable, neutral, and unstable stratification. The impact of thermal stratification on turbulent flows over a steep three-dimensional hill is clarified through flow patterns and statistical characteristics. Compared to neutral stratification, the separation bubble downstream of the hill crest is reduced under unstable stratification, while it is enlarged under stable stratification. In addition, turbulent eddy motions in the wake region are enhanced in the unstable condition but are suppressed in the stable condition. Both mean velocities and turbulence fluctuations over steep hilly terrain are amplified by unstable stratification and attenuated by stable stratification. The flow characteristics on the hill crest are comprehensively determined by the topography and thermal stratification, whereas the flow dynamics in the hill wake are predominantly influenced by terrain-induced turbulence. Moreover, the mechanisms driving the formation of flow fields over steep hilly topography under different thermal stratification are investigated through force balance analysis using the time-averaged Navier–Stokes equations. The results indicate that turbulence plays a negligible role in the force balance upstream of the hill, while it becomes the dominant factor for the force balances downstream of the hill.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106521"},"PeriodicalIF":2.5,"publicationDate":"2024-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138652","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A FEM computational approach for gas-liquid flow in pipelines using a two-fluid model

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-12-06 DOI: 10.1016/j.compfluid.2024.106520

Xiaowei Li , Ruichao Tian , Limin He , Yuling Lv , Shidong Zhou , Yaqiang Li

Two-phase flow is typically observed in gas-liquid pipelines across diverse domains, including nuclear, petroleum, and chemical industries. As accurate prediction of flow characteristics is crucial for engineering applications, one-dimensional two-fluid models with various treatments have been extensively employed to mathematically describe the gas-liquid variations in the pipelines through a set of non-linear partial differential equations (PDEs). This paper presents a modularly designed algorithm that incorporates an implicit scheme coupled with the finite element method (FEM) to solve the one-dimensional two-fluid model with gas-liquid stratified calculation. To validate the accuracy of this algorithm, four cases utilizing varying mesh sizes, inlet flows, and outlet pressures are conducted to scrutinize numerical steady-state gas-liquid flow characteristics, and the consistency between the numerical variations computed through this algorithm and those from OLGA simulator is used to analyze transient gas-liquid behaviors. The steady-state flow fields reveal two distinct zones along the pipe: an intense momentum exchange zone influenced by the inlet nonequilibrium state and a gentle momentum exchange zone influenced by the gas compressibility. Notably, a finer mesh will yield more accurate descriptions of flow parameters in the intense zone, while a relatively sparser mesh suffices for the gentle zone. Additionally, the transient results reveal that the gas-liquid variations in the pipe under initial condition of single-phase gas can be divided into three stages: the gas expansion stage determined by gas compressibility, the gas spread stage influenced by the gas propulsion, and the liquid filling stage decided by the liquid kinetic motion. The consistent identification of the three stages in gas-liquid variations under initial conditions of different static fluids highlights the effectiveness and accuracy of the proposed numerical method in describing transient features.

{"title":"A FEM computational approach for gas-liquid flow in pipelines using a two-fluid model","authors":"Xiaowei Li , Ruichao Tian , Limin He , Yuling Lv , Shidong Zhou , Yaqiang Li","doi":"10.1016/j.compfluid.2024.106520","DOIUrl":"10.1016/j.compfluid.2024.106520","url":null,"abstract":"<div><div>Two-phase flow is typically observed in gas-liquid pipelines across diverse domains, including nuclear, petroleum, and chemical industries. As accurate prediction of flow characteristics is crucial for engineering applications, one-dimensional two-fluid models with various treatments have been extensively employed to mathematically describe the gas-liquid variations in the pipelines through a set of non-linear partial differential equations (PDEs). This paper presents a modularly designed algorithm that incorporates an implicit scheme coupled with the finite element method (FEM) to solve the one-dimensional two-fluid model with gas-liquid stratified calculation. To validate the accuracy of this algorithm, four cases utilizing varying mesh sizes, inlet flows, and outlet pressures are conducted to scrutinize numerical steady-state gas-liquid flow characteristics, and the consistency between the numerical variations computed through this algorithm and those from OLGA simulator is used to analyze transient gas-liquid behaviors. The steady-state flow fields reveal two distinct zones along the pipe: an intense momentum exchange zone influenced by the inlet nonequilibrium state and a gentle momentum exchange zone influenced by the gas compressibility. Notably, a finer mesh will yield more accurate descriptions of flow parameters in the intense zone, while a relatively sparser mesh suffices for the gentle zone. Additionally, the transient results reveal that the gas-liquid variations in the pipe under initial condition of single-phase gas can be divided into three stages: the gas expansion stage determined by gas compressibility, the gas spread stage influenced by the gas propulsion, and the liquid filling stage decided by the liquid kinetic motion. The consistent identification of the three stages in gas-liquid variations under initial conditions of different static fluids highlights the effectiveness and accuracy of the proposed numerical method in describing transient features.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106520"},"PeriodicalIF":2.5,"publicationDate":"2024-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138566","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Parallel large eddy simulations with curvilinear immersed boundary method for high-speed flows

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-12-06 DOI: 10.1016/j.compfluid.2024.106495

Amir M. Akbarzadeh, Iman Borazjani

A sharp-interface immersed boundary method is developed to simulate turbulent compressible flows through large-eddy simulations (LES) on curvilinear grids. The curvilinear grid enables increasing the grid resolution near regions of interest such as solid walls. To capture both shocks and turbulence, the equations are discretized using a hybrid discretization comprising a fourth-order skew-central scheme and a third-order weighted essentially nonoscillatory (WENO) scheme. A switch function is incorporated to switch between WENO in the vicinity of shocks to central far away from shocks. A dynamics Smagornisky model is used to model the subgrid scales for the central scheme. The interpolation for the immersed boundary is modified to incorporate wall functions. The code is parallelized to efficiently run on thousands of CPU cores for highly resolved grids. The method is verified and validated against several test cases including a decaying isotropic turbulent flow, turbulent channel flow, supersonic flow and shock diffraction over a cylinder. The results show that the LES can properly resolve the inertial subrange and the hybrid scheme can effectively capture shocks over the immersed bodies. It is observed that highly refined grids and low-dissipation hybrid scheme are necessary to capture fine turbulence features such as shear instabilities and shock boundary layer interaction over immersed bodies. In fine grids, however, the importance of explicit LES modeling decreases as most scales are resolved and the WENO scheme provides the dissipation implicitly. In such cases, the results are most sensitive to wall modeling which demonstrate the need for development of wall models for high-speed flows.

{"title":"Parallel large eddy simulations with curvilinear immersed boundary method for high-speed flows","authors":"Amir M. Akbarzadeh, Iman Borazjani","doi":"10.1016/j.compfluid.2024.106495","DOIUrl":"10.1016/j.compfluid.2024.106495","url":null,"abstract":"<div><div>A sharp-interface immersed boundary method is developed to simulate turbulent compressible flows through large-eddy simulations (LES) on curvilinear grids. The curvilinear grid enables increasing the grid resolution near regions of interest such as solid walls. To capture both shocks and turbulence, the equations are discretized using a hybrid discretization comprising a fourth-order skew-central scheme and a third-order weighted essentially nonoscillatory (WENO) scheme. A switch function is incorporated to switch between WENO in the vicinity of shocks to central far away from shocks. A dynamics Smagornisky model is used to model the subgrid scales for the central scheme. The interpolation for the immersed boundary is modified to incorporate wall functions. The code is parallelized to efficiently run on thousands of CPU cores for highly resolved grids. The method is verified and validated against several test cases including a decaying isotropic turbulent flow, turbulent channel flow, supersonic flow and shock diffraction over a cylinder. The results show that the LES can properly resolve the inertial subrange and the hybrid scheme can effectively capture shocks over the immersed bodies. It is observed that highly refined grids and low-dissipation hybrid scheme are necessary to capture fine turbulence features such as shear instabilities and shock boundary layer interaction over immersed bodies. In fine grids, however, the importance of explicit LES modeling decreases as most scales are resolved and the WENO scheme provides the dissipation implicitly. In such cases, the results are most sensitive to wall modeling which demonstrate the need for development of wall models for high-speed flows.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106495"},"PeriodicalIF":2.5,"publicationDate":"2024-12-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138565","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A heterogeneous hybrid-precision finite volume method for compressible flow on unstructured grids

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-12-04 DOI: 10.1016/j.compfluid.2024.106505

Chen Wang, Jian Xia, Long Chen

Single-precision floating-point GPU calculations in modern high-performance heterogeneous computing systems are crucial for increasing the efficiency of large-scale fluid simulations on unstructured grids. However, the lack of a unified programming language for heterogeneous systems and the significant computational errors of single-precision calculations in complex problems pose major challenges. Issues such as poor data locality and data contention in unstructured grid CFD calculations limit GPU performance. Through heterogeneous Kokkos computation, we improved data locality through data reordering and addressed data contention using the scatter-reduce strategy, atomic operations, and the color approach. We introduced an innovative hybrid-precision CFD computation strategy that leverages methods based on object distance and grid geometry for precision blending. This approach harnesses the computational advantages of single-precision GPU calculations while accurately capturing boundary layer information. We assessed the accuracy and performance of these methods on a heterogeneous CPU/GPU computing system. The reverse Cuthill-McKee algorithm significantly enhances performance, atomic operations are the optimal strategy for GPUs, and in the hybrid-precision strategy proposed in this paper, the Tesla A100 GPU, RTX 4090 GPU, and RX 7900 XTX GPU achieve overall speedup of 469, 310, and 413, respectively.

{"title":"A heterogeneous hybrid-precision finite volume method for compressible flow on unstructured grids","authors":"Chen Wang, Jian Xia, Long Chen","doi":"10.1016/j.compfluid.2024.106505","DOIUrl":"10.1016/j.compfluid.2024.106505","url":null,"abstract":"<div><div>Single-precision floating-point GPU calculations in modern high-performance heterogeneous computing systems are crucial for increasing the efficiency of large-scale fluid simulations on unstructured grids. However, the lack of a unified programming language for heterogeneous systems and the significant computational errors of single-precision calculations in complex problems pose major challenges. Issues such as poor data locality and data contention in unstructured grid CFD calculations limit GPU performance. Through heterogeneous Kokkos computation, we improved data locality through data reordering and addressed data contention using the scatter-reduce strategy, atomic operations, and the color approach. We introduced an innovative hybrid-precision CFD computation strategy that leverages methods based on object distance and grid geometry for precision blending. This approach harnesses the computational advantages of single-precision GPU calculations while accurately capturing boundary layer information. We assessed the accuracy and performance of these methods on a heterogeneous CPU/GPU computing system. The reverse Cuthill-McKee algorithm significantly enhances performance, atomic operations are the optimal strategy for GPUs, and in the hybrid-precision strategy proposed in this paper, the Tesla A100 GPU, RTX 4090 GPU, and RX 7900 XTX GPU achieve overall speedup of 469, 310, and 413, respectively.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106505"},"PeriodicalIF":2.5,"publicationDate":"2024-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138563","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Using Delayed Detached Eddy Simulation to create datasets for data-driven turbulence modeling: A periodic hills with parameterized geometry case

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-12-04 DOI: 10.1016/j.compfluid.2024.106506

Davide Oberto , Davide Fransos , Stefano Berrone

Despite the emerging field of data-driven turbulence models, there is a lack of systematic high-fidelity datasets at flow configurations changing continuously with respect to geometrical/physical parameters. In this work, we investigate the possibility of using Delayed Detached Eddy Simulation (DDES) to generate reliable datasets in a significantly cheaper manner compared to the DNS or LES counterparts. To do that, we perform 25 simulations of the geometrically-parameterized periodic hills test case to deal with different hills steepnesses. We firstly check the accuracy of our results by comparing one simulation with the benchmark case of Xiao et al. Then, we use such database to train the turbulent viscosity-Vector Basis Neural Network (

ν_{t}

-VBNN) data-driven turbulence model. The latter outperforms the classic

k - ω

SST RANS model, proving that our generated dataset can be useful for data-driven turbulence modeling and opening the opportunity to exploit DDES to create systematic datasets for data-driven turbulence modeling.

{"title":"Using Delayed Detached Eddy Simulation to create datasets for data-driven turbulence modeling: A periodic hills with parameterized geometry case","authors":"Davide Oberto , Davide Fransos , Stefano Berrone","doi":"10.1016/j.compfluid.2024.106506","DOIUrl":"10.1016/j.compfluid.2024.106506","url":null,"abstract":"<div><div>Despite the emerging field of data-driven turbulence models, there is a lack of systematic high-fidelity datasets at flow configurations changing continuously with respect to geometrical/physical parameters. In this work, we investigate the possibility of using Delayed Detached Eddy Simulation (DDES) to generate reliable datasets in a significantly cheaper manner compared to the DNS or LES counterparts. To do that, we perform 25 simulations of the geometrically-parameterized periodic hills test case to deal with different hills steepnesses. We firstly check the accuracy of our results by comparing one simulation with the benchmark case of Xiao et al. Then, we use such database to train the turbulent viscosity-Vector Basis Neural Network (<span><math><msub><mrow><mi>ν</mi></mrow><mrow><mi>t</mi></mrow></msub></math></span>-VBNN) data-driven turbulence model. The latter outperforms the classic <span><math><mrow><mi>k</mi><mo>−</mo><mi>ω</mi></mrow></math></span> SST RANS model, proving that our generated dataset can be useful for data-driven turbulence modeling and opening the opportunity to exploit DDES to create systematic datasets for data-driven turbulence modeling.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106506"},"PeriodicalIF":2.5,"publicationDate":"2024-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138731","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Parallel implementation and performance of super-resolution generative adversarial network turbulence models for large-eddy simulation

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-12-02 DOI: 10.1016/j.compfluid.2024.106498

Ludovico Nista , Christoph D.K. Schumann , Peicho Petkov , Valentin Pavlov , Temistocle Grenga , Jonathan F. MacArt , Antonio Attili , Stoyan Markov , Heinz Pitsch

Super-resolution (SR) generative adversarial networks (GANs) are promising for turbulence closure in large-eddy simulation (LES) due to their ability to accurately reconstruct high-resolution data from low-resolution fields. Current model training and inference strategies are not sufficiently mature for large-scale, distributed calculations due to the computational demands and often unstable training of SR-GANs, which limits the exploration of improved model structures, training strategies, and loss-function definitions. Integrating SR-GANs into LES solvers for inference-coupled simulations is also necessary to assess their a posteriori accuracy, stability, and cost. We investigate parallelization strategies for SR-GAN training and inference-coupled LES, focusing on computational performance and reconstruction accuracy. We examine distributed data-parallel training strategies for hybrid CPU–GPU node architectures and the associated influence of low-/high-resolution subbox size, global batch size, and discriminator accuracy. Accurate predictions require training subboxes that are sufficiently large relative to the Kolmogorov length scale. Care should be placed on the coupled effect of training batch size, learning rate, number of training subboxes, and discriminator’s learning capabilities. We introduce a data-parallel SR-GAN training and inference library for heterogeneous architectures that enables exchange between the LES solver and SR-GAN inference at runtime. We investigate the predictive accuracy and computational performance of this arrangement with particular focus on the overlap (halo) size required for accurate SR reconstruction. Similarly, a posteriori parallel scaling for efficient inference-coupled LES is constrained by the SR subdomain size, GPU utilization, and reconstruction accuracy. Based on these findings, we establish guidelines and best practices to optimize resource utilization and parallel acceleration of SR-GAN turbulence model training and inference-coupled LES calculations while maintaining predictive accuracy.

{"title":"Parallel implementation and performance of super-resolution generative adversarial network turbulence models for large-eddy simulation","authors":"Ludovico Nista , Christoph D.K. Schumann , Peicho Petkov , Valentin Pavlov , Temistocle Grenga , Jonathan F. MacArt , Antonio Attili , Stoyan Markov , Heinz Pitsch","doi":"10.1016/j.compfluid.2024.106498","DOIUrl":"10.1016/j.compfluid.2024.106498","url":null,"abstract":"<div><div>Super-resolution (SR) generative adversarial networks (GANs) are promising for turbulence closure in large-eddy simulation (LES) due to their ability to accurately reconstruct high-resolution data from low-resolution fields. Current model training and inference strategies are not sufficiently mature for large-scale, distributed calculations due to the computational demands and often unstable training of SR-GANs, which limits the exploration of improved model structures, training strategies, and loss-function definitions. Integrating SR-GANs into LES solvers for inference-coupled simulations is also necessary to assess their <em>a posteriori</em> accuracy, stability, and cost. We investigate parallelization strategies for SR-GAN training and inference-coupled LES, focusing on computational performance and reconstruction accuracy. We examine distributed data-parallel training strategies for hybrid CPU–GPU node architectures and the associated influence of low-/high-resolution subbox size, global batch size, and discriminator accuracy. Accurate predictions require training subboxes that are sufficiently large relative to the Kolmogorov length scale. Care should be placed on the coupled effect of training batch size, learning rate, number of training subboxes, and discriminator’s learning capabilities. We introduce a data-parallel SR-GAN training and inference library for heterogeneous architectures that enables exchange between the LES solver and SR-GAN inference at runtime. We investigate the predictive accuracy and computational performance of this arrangement with particular focus on the overlap (halo) size required for accurate SR reconstruction. Similarly, <em>a posteriori</em> parallel scaling for efficient inference-coupled LES is constrained by the SR subdomain size, GPU utilization, and reconstruction accuracy. Based on these findings, we establish guidelines and best practices to optimize resource utilization and parallel acceleration of SR-GAN turbulence model training and inference-coupled LES calculations while maintaining predictive accuracy.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106498"},"PeriodicalIF":2.5,"publicationDate":"2024-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138560","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Optimized dynamic similarity models to predict SGS backscatter in 2D decaying turbulence

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-11-30 DOI: 10.1016/j.compfluid.2024.106497

Dandan Wang , Yu-xin Ren , Mengnan Ding

Large eddy simulation (LES) of two-dimensional (2D) turbulence is often used in the geostrophic flows. However, some basic dynamics underlying traditional SGS models are absent in 2D turbulence, e.g. the vortex stretching. Hence, this research proposes an optimized dynamic similarity model (DSM) for the SGS stress, which is constructed through the dynamic procedure based on the Germano identity. In addition, a modification is made to the dynamic mixed model (DMM) for the sake of realizability condition. The optimized DSM is justified in comparison with the DMM, through the a priori and a posteriori verifications, in the context of the 2D decaying turbulence with turbulent Reynolds number of

R e = 3.7 \times 1 0^{4}

and turbulent Mach number of

M_{t} = 0.1

. Special attention is paid to the consistency of the verification procedure, so that the filtering operations used in the direct numerical simulation (DNS) and LES are optimally equivalent. The SGS transport phenomena, especially the SGS backscatter, predicted by these two models are studied in detail. In addition, the optimized DSM and the DMM are extended for the modified SGS transport vectors of passive scalars to show their capability in calculating 2D turbulent mixing. The numerical results show the optimized DSM provides larger correlation coefficient, better locality, and stronger SGS backscsatter than the DMM does, and therefore it is more suitable for the LES of 2D turbulence.

{"title":"Optimized dynamic similarity models to predict SGS backscatter in 2D decaying turbulence","authors":"Dandan Wang , Yu-xin Ren , Mengnan Ding","doi":"10.1016/j.compfluid.2024.106497","DOIUrl":"10.1016/j.compfluid.2024.106497","url":null,"abstract":"<div><div>Large eddy simulation (LES) of two-dimensional (2D) turbulence is often used in the geostrophic flows. However, some basic dynamics underlying traditional SGS models are absent in 2D turbulence, e.g. the vortex stretching. Hence, this research proposes an optimized dynamic similarity model (DSM) for the SGS stress, which is constructed through the dynamic procedure based on the Germano identity. In addition, a modification is made to the dynamic mixed model (DMM) for the sake of realizability condition. The optimized DSM is justified in comparison with the DMM, through the a priori and a posteriori verifications, in the context of the 2D decaying turbulence with turbulent Reynolds number of <span><math><mrow><mi>R</mi><mi>e</mi><mo>=</mo><mn>3</mn><mo>.</mo><mn>7</mn><mo>×</mo><mn>1</mn><msup><mrow><mn>0</mn></mrow><mrow><mn>4</mn></mrow></msup></mrow></math></span> and turbulent Mach number of <span><math><mrow><msub><mrow><mi>M</mi></mrow><mrow><mi>t</mi></mrow></msub><mo>=</mo><mn>0</mn><mo>.</mo><mn>1</mn></mrow></math></span>. Special attention is paid to the consistency of the verification procedure, so that the filtering operations used in the direct numerical simulation (DNS) and LES are optimally equivalent. The SGS transport phenomena, especially the SGS backscatter, predicted by these two models are studied in detail. In addition, the optimized DSM and the DMM are extended for the modified SGS transport vectors of passive scalars to show their capability in calculating 2D turbulent mixing. The numerical results show the optimized DSM provides larger correlation coefficient, better locality, and stronger SGS backscsatter than the DMM does, and therefore it is more suitable for the LES of 2D turbulence.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106497"},"PeriodicalIF":2.5,"publicationDate":"2024-11-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138562","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Direct numerical simulations of two-dimensional channel flow with a gap deformity and slip wall

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-11-30 DOI: 10.1016/j.compfluid.2024.106496

Silvia Ceccacci , Sophie A.W. Calabretto , Christian Thomas , James P. Denier

The effect of surface slip on the dynamics of flow separation induced by a Gaussian-shaped gap deformity in a two-dimensional channel was numerically investigated for Reynolds numbers

Re \in [100, 6000]

. Two gap deformations, denoted wide and narrow, were modelled with dimensions sufficient to generate localised pockets of reversed flow when the channel walls were fully no-slip. The wide gap induces a more intense region of separated flow than the narrow gap but less than that exhibited by similar-sized bumps in a channel (Ceccacci et al., 2022). In addition, the size and magnitude of the separation bubble within each gap deformity plateaued for Reynolds numbers

Re > 3000

. Surface slip with slip length,

λ

, was modelled via a Navier-slip boundary condition. Applying the slip condition to the gap concavity reduces the magnitude and thickness of the separation bubble within the deformation and, for a slip length

λ \approx 0.1

, eliminates separated flow for both gap configurations, which is less than the requirements for the bump configuration (Ceccacci et al., 2022). Moreover, limiting slip to the gap region, achieved the same flow separation control, as that realised by applying slip to the entire wall.

{"title":"Direct numerical simulations of two-dimensional channel flow with a gap deformity and slip wall","authors":"Silvia Ceccacci , Sophie A.W. Calabretto , Christian Thomas , James P. Denier","doi":"10.1016/j.compfluid.2024.106496","DOIUrl":"10.1016/j.compfluid.2024.106496","url":null,"abstract":"<div><div>The effect of surface slip on the dynamics of flow separation induced by a Gaussian-shaped gap deformity in a two-dimensional channel was numerically investigated for Reynolds numbers <span><math><mrow><mi>Re</mi><mo>∈</mo><mrow><mo>[</mo><mn>100</mn><mo>,</mo><mn>6000</mn><mo>]</mo></mrow></mrow></math></span>. Two gap deformations, denoted wide and narrow, were modelled with dimensions sufficient to generate localised pockets of reversed flow when the channel walls were fully no-slip. The wide gap induces a more intense region of separated flow than the narrow gap but less than that exhibited by similar-sized bumps in a channel (Ceccacci et al., 2022). In addition, the size and magnitude of the separation bubble within each gap deformity plateaued for Reynolds numbers <span><math><mrow><mi>Re</mi><mo>></mo><mn>3000</mn></mrow></math></span>. Surface slip with slip length, <span><math><mi>λ</mi></math></span>, was modelled via a Navier-slip boundary condition. Applying the slip condition to the gap concavity reduces the magnitude and thickness of the separation bubble within the deformation and, for a slip length <span><math><mrow><mi>λ</mi><mo>≈</mo><mn>0</mn><mo>.</mo><mn>1</mn></mrow></math></span>, eliminates separated flow for both gap configurations, which is less than the requirements for the bump configuration (Ceccacci et al., 2022). Moreover, limiting slip to the gap region, achieved the same flow separation control, as that realised by applying slip to the entire wall.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106496"},"PeriodicalIF":2.5,"publicationDate":"2024-11-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138729","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Investigation of aerodynamic performance and noise of tip shape clearance in a diagonal flow fan

IF 2.5 3区工程技术 Q3 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Fluids

Pub Date : 2024-11-30 DOI: 10.1016/j.compfluid.2024.106471

A. Zijian Mao (毛子鉴) , B. Shuiqing Zhou (周水清) , C. Tianle Zhang (张天乐) , D. Jiacheng He (何嘉成) , E. Weiya Jin (金伟娅) , F. Weiping Feng (冯伟平)

Unlike centrifugal fans, diagonal flow fans have an air inflow direction at a certain angle to the axis, with tip clearance (TC) significantly affecting aerodynamic performance and noise. This study focuses on investigating the effects of four different TC shapes based on blade height (BH) ratios. Large Eddy Simulation (LES) was used to capture the structure and development of tip leakage flow (TLF) and tip leakage vortex (TLV) accurately. Diagonal flow fan casings with different TC shapes were 3D-printed for aerodynamic performance experiments. The results from both numerical simulations and experiments show that diagonal flow fans with Tapering or Parallel TC shapes achieve superior aerodynamic performance compared to other geometries, with a 24.1% variance in flow rate. The study further indicates that different TC shapes significantly influence the flow field, altering the mechanisms governing turbulence transition on the suction side. Compared to the Diverging shape, reducing the TC width decreases the amplitude of the TLV, which in turn reduces the turbulence-affected area while increasing dominant mode frequencies. Experimental results also confirm that the Tapering-Diverging TC shape yields the lowest noise levels, with a 3.6 dB reduction in Sound Pressure Level (SPL).

{"title":"Investigation of aerodynamic performance and noise of tip shape clearance in a diagonal flow fan","authors":"A. Zijian Mao (毛子鉴) , B. Shuiqing Zhou (周水清) , C. Tianle Zhang (张天乐) , D. Jiacheng He (何嘉成) , E. Weiya Jin (金伟娅) , F. Weiping Feng (冯伟平)","doi":"10.1016/j.compfluid.2024.106471","DOIUrl":"10.1016/j.compfluid.2024.106471","url":null,"abstract":"<div><div>Unlike centrifugal fans, diagonal flow fans have an air inflow direction at a certain angle to the axis, with tip clearance (TC) significantly affecting aerodynamic performance and noise. This study focuses on investigating the effects of four different TC shapes based on blade height (BH) ratios. Large Eddy Simulation (LES) was used to capture the structure and development of tip leakage flow (TLF) and tip leakage vortex (TLV) accurately. Diagonal flow fan casings with different TC shapes were 3D-printed for aerodynamic performance experiments. The results from both numerical simulations and experiments show that diagonal flow fans with Tapering or Parallel TC shapes achieve superior aerodynamic performance compared to other geometries, with a 24.1% variance in flow rate. The study further indicates that different TC shapes significantly influence the flow field, altering the mechanisms governing turbulence transition on the suction side. Compared to the Diverging shape, reducing the TC width decreases the amplitude of the TLV, which in turn reduces the turbulence-affected area while increasing dominant mode frequencies. Experimental results also confirm that the Tapering-Diverging TC shape yields the lowest noise levels, with a 3.6 dB reduction in Sound Pressure Level (SPL).</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"288 ","pages":"Article 106471"},"PeriodicalIF":2.5,"publicationDate":"2024-11-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143138564","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0