Fenghui Lin , Zi-Mo Liao , Zhiye Zhao , Nansheng Liu , Xi-Yun Lu , Bamin Khomami
{"title":"用于聚合物诱导/修正湍流直接数值模拟的高保真算法的GPU加速","authors":"Fenghui Lin , Zi-Mo Liao , Zhiye Zhao , Nansheng Liu , Xi-Yun Lu , Bamin Khomami","doi":"10.1016/j.jnnfm.2025.105437","DOIUrl":null,"url":null,"abstract":"<div><div>This paper presents an efficient GPU implementation for the direct numerical simulation (DNS) of polymer-induced/modified turbulence, utilizing the open-source finite-difference incompressible Navier–Stokes solver, <span><math><mrow><mi>C</mi><mi>a</mi><mi>N</mi><mi>S</mi></mrow></math></span>. The implementation incorporates a versatile viscoelastic solver for two commonly used constitutive equations in DNS of elastic or elastically modified turbulence, namely, the FENE-P and the Giesekus models. Consistent with <span><math><mrow><mi>C</mi><mi>a</mi><mi>N</mi><mi>S</mi></mrow></math></span>, the viscoelastic solver uses CUDA Fortran and makes extensive use of kernel loop directives (CUF kernels). To improve the fidelity and robustness of this implementation, a tensor-based interpolation method combined with a shock-capturing WENO scheme is employed for spatial discretization of the polymer constitutive equations. We demonstrate the accuracy and robustness of our code by comparison with existing theoretical and simulation results. In addition, the algorithm exhibits superior scalability with up to eight Nvidia GPU devices in benchmark channel flows. In turn, we use this expeditious code for high-fidelity and efficient large-scale DNS of viscoelastic turbulent flows. To that end, we demonstrate the broad applicability of our implementation in a host of polymer-induced/modified turbulence in channel flows. This includes the maximum drag reduction asymptote, as well as elasto-inertial turbulence, and purely elastic turbulent flows. Overall, this GPU-accelerated simulation technique has all the required ingredients to become the method of choice for large-scale viscoelastic computations aimed at faithfully capturing polymer-induced flow phenomena.</div></div>","PeriodicalId":54782,"journal":{"name":"Journal of Non-Newtonian Fluid Mechanics","volume":"342 ","pages":"Article 105437"},"PeriodicalIF":2.8000,"publicationDate":"2025-06-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"GPU acceleration of a hi-fidelity algorithm for direct numerical simulation of polymer-induced/modified turbulence\",\"authors\":\"Fenghui Lin , Zi-Mo Liao , Zhiye Zhao , Nansheng Liu , Xi-Yun Lu , Bamin Khomami\",\"doi\":\"10.1016/j.jnnfm.2025.105437\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>This paper presents an efficient GPU implementation for the direct numerical simulation (DNS) of polymer-induced/modified turbulence, utilizing the open-source finite-difference incompressible Navier–Stokes solver, <span><math><mrow><mi>C</mi><mi>a</mi><mi>N</mi><mi>S</mi></mrow></math></span>. The implementation incorporates a versatile viscoelastic solver for two commonly used constitutive equations in DNS of elastic or elastically modified turbulence, namely, the FENE-P and the Giesekus models. Consistent with <span><math><mrow><mi>C</mi><mi>a</mi><mi>N</mi><mi>S</mi></mrow></math></span>, the viscoelastic solver uses CUDA Fortran and makes extensive use of kernel loop directives (CUF kernels). To improve the fidelity and robustness of this implementation, a tensor-based interpolation method combined with a shock-capturing WENO scheme is employed for spatial discretization of the polymer constitutive equations. We demonstrate the accuracy and robustness of our code by comparison with existing theoretical and simulation results. In addition, the algorithm exhibits superior scalability with up to eight Nvidia GPU devices in benchmark channel flows. In turn, we use this expeditious code for high-fidelity and efficient large-scale DNS of viscoelastic turbulent flows. To that end, we demonstrate the broad applicability of our implementation in a host of polymer-induced/modified turbulence in channel flows. This includes the maximum drag reduction asymptote, as well as elasto-inertial turbulence, and purely elastic turbulent flows. Overall, this GPU-accelerated simulation technique has all the required ingredients to become the method of choice for large-scale viscoelastic computations aimed at faithfully capturing polymer-induced flow phenomena.</div></div>\",\"PeriodicalId\":54782,\"journal\":{\"name\":\"Journal of Non-Newtonian Fluid Mechanics\",\"volume\":\"342 \",\"pages\":\"Article 105437\"},\"PeriodicalIF\":2.8000,\"publicationDate\":\"2025-06-03\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of Non-Newtonian Fluid Mechanics\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S0377025725000564\",\"RegionNum\":2,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"MECHANICS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Non-Newtonian Fluid Mechanics","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0377025725000564","RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"MECHANICS","Score":null,"Total":0}
GPU acceleration of a hi-fidelity algorithm for direct numerical simulation of polymer-induced/modified turbulence
This paper presents an efficient GPU implementation for the direct numerical simulation (DNS) of polymer-induced/modified turbulence, utilizing the open-source finite-difference incompressible Navier–Stokes solver, . The implementation incorporates a versatile viscoelastic solver for two commonly used constitutive equations in DNS of elastic or elastically modified turbulence, namely, the FENE-P and the Giesekus models. Consistent with , the viscoelastic solver uses CUDA Fortran and makes extensive use of kernel loop directives (CUF kernels). To improve the fidelity and robustness of this implementation, a tensor-based interpolation method combined with a shock-capturing WENO scheme is employed for spatial discretization of the polymer constitutive equations. We demonstrate the accuracy and robustness of our code by comparison with existing theoretical and simulation results. In addition, the algorithm exhibits superior scalability with up to eight Nvidia GPU devices in benchmark channel flows. In turn, we use this expeditious code for high-fidelity and efficient large-scale DNS of viscoelastic turbulent flows. To that end, we demonstrate the broad applicability of our implementation in a host of polymer-induced/modified turbulence in channel flows. This includes the maximum drag reduction asymptote, as well as elasto-inertial turbulence, and purely elastic turbulent flows. Overall, this GPU-accelerated simulation technique has all the required ingredients to become the method of choice for large-scale viscoelastic computations aimed at faithfully capturing polymer-induced flow phenomena.
期刊介绍:
The Journal of Non-Newtonian Fluid Mechanics publishes research on flowing soft matter systems. Submissions in all areas of flowing complex fluids are welcomed, including polymer melts and solutions, suspensions, colloids, surfactant solutions, biological fluids, gels, liquid crystals and granular materials. Flow problems relevant to microfluidics, lab-on-a-chip, nanofluidics, biological flows, geophysical flows, industrial processes and other applications are of interest.
Subjects considered suitable for the journal include the following (not necessarily in order of importance):
Theoretical, computational and experimental studies of naturally or technologically relevant flow problems where the non-Newtonian nature of the fluid is important in determining the character of the flow. We seek in particular studies that lend mechanistic insight into flow behavior in complex fluids or highlight flow phenomena unique to complex fluids. Examples include
Instabilities, unsteady and turbulent or chaotic flow characteristics in non-Newtonian fluids,
Multiphase flows involving complex fluids,
Problems involving transport phenomena such as heat and mass transfer and mixing, to the extent that the non-Newtonian flow behavior is central to the transport phenomena,
Novel flow situations that suggest the need for further theoretical study,
Practical situations of flow that are in need of systematic theoretical and experimental research. Such issues and developments commonly arise, for example, in the polymer processing, petroleum, pharmaceutical, biomedical and consumer product industries.