Configuring computational fluid dynamics(CFD)simulations typically demands extensive domain expertise,limiting broader access.Although large language models(LLMs)have advanced scientific computing,their use in automat...Configuring computational fluid dynamics(CFD)simulations typically demands extensive domain expertise,limiting broader access.Although large language models(LLMs)have advanced scientific computing,their use in automating CFD workflows is underdeveloped.We introduce a novel approach centered on domain-specific LLM adaptation.By fine-tuning Qwen2.5-7B-Instruct on NL2FOAM,our custom dataset of 28,716 natural language-to-OpenFOAM configuration pairs with chain-of-thought(CoT)annotations enables direct translation from natural language descriptions to executable CFD setups.A multi-agent system orchestrates the process,autonomously verifying inputs,generating configurations,running simulations,and correcting errors.Evaluation on a benchmark of 21 diverse flow cases demonstrates state-of-the-art performance,achieving 88.7%solution accuracy and 82.6%first-attempt success rate.This significantly outperforms larger general-purpose models such as Qwen2.5-72B-Instruct,DeepSeek-R1,and Llama3.3-70B-Instruct,while also requiring fewer correction iterations and maintaining high computational efficiency.The results highlight the critical role of domain-specific adaptation in deploying LLM assistants for complex engineering workflows.Our code and fine-tuned model have been deposited at https://github.com/YYgroup/AutoCFD.展开更多
We develop and assess a model of the turbulent burning velocity ST over a wide range of conditions.The aim is to obtain an explicit ST model for turbulent combustion modeling and flame analysis.The model consists of s...We develop and assess a model of the turbulent burning velocity ST over a wide range of conditions.The aim is to obtain an explicit ST model for turbulent combustion modeling and flame analysis.The model consists of sub models of the stretch factor and the turbulent flame area.The stretch factor characterizes the flame response of turbulence stretch and incorporates detailed chemistry and transport effects with a lookup table of laminar counterflow flames.The flame area model captures the area growth based on Lagrangian statistics of propagating surfaces and considers the effects of turbulence length scales and fuel characteristics.The present model predicts sT via an algebraic expression without free parameters.We assess the model using 490 cases of the direct numerical simulation or experiment reported from various research groups on planar and Bunsen flames over a wide range of conditions,covering fuels from hydrogen to n-dodecane,pressures from 1 to 30 atm,lean and rich mixtures,turbulence intensity ratios from 0.1 to 177.6,and turbulence length ratios from 0.5 to 66.7.Despite the scattering sT data in the literature,the comprehensive comparison shows that the proposed ST model has an overall good agreement over the wide range of conditions,with the averaged modeling error of 28.1%.展开更多
We propose a new flame index for the transported probability density function(PDF) method. The flame index uses mixing flux projections of Lagrangian particles on mixture fraction and progress variable directions as t...We propose a new flame index for the transported probability density function(PDF) method. The flame index uses mixing flux projections of Lagrangian particles on mixture fraction and progress variable directions as the metrics to identify the combustion mode, with the Burke-Schumann solution as a reference. A priori validation of the flame index is conducted with a series of constructed turbulent partially premixed reactors. It indicates that the proposed flame index is able to identify the combustion mode based on the subgrid mixing information. The flame index is then applied the large eddy simulation/PDF datasets of turbulent partially premixed jet flames. Results show that the flame index separate different combustion modes and extinction correctly. The proposed flame index provides a promising tool to analyze and model the partially premixed flames adaptively.展开更多
Quantum computing has grown substantially over the past four decades,but whether it can outperform classical methods in practical use remains uncertain[1].Fluid dynamics simulation,challenging in classical physics but...Quantum computing has grown substantially over the past four decades,but whether it can outperform classical methods in practical use remains uncertain[1].Fluid dynamics simulation,challenging in classical physics but vital for applications,is a potential area for showcasing quantum advantage.The quantum computing for fluid dynamics(QCFD)[2]is expected to efficiently simulate intricate turbulent flows with high Reynolds numbers.This capability is crucial for critical applications,including aircraft design and weather forecast.展开更多
Bio-inspired micro-air-vehicles(MAVs)usually operate in the atmospheric boundary layer at a low Reynolds number and complex wind conditions including large-scale turbulence,strong shear,and gusts.We develop an open je...Bio-inspired micro-air-vehicles(MAVs)usually operate in the atmospheric boundary layer at a low Reynolds number and complex wind conditions including large-scale turbulence,strong shear,and gusts.We develop an open jet facility(OJF)to meet the requirements of MAV flight experiments at very low speed and high turbulence intensity.Powered by a stage-driven fan,the OJF is capable of generating wind speeds covering 0.1–16.8 m/s,with a velocity ratio of 100:1.The contraction section of the OJF is designed using an adjoint-driven optimization method,resulting in a con-traction ratio of 3:1 and a length-to-diameter ratio of 0.75.A modularized design of the jet nozzle can produce laminar or high-turbulence wind conditions.Flow field calibration results demonstrate that the OJF is capable of producing a high-quality baseline flow with steady airspeed as low as 0.1 m/s,uniform region around 80%of the cross-sectional test area,and turbulence intensity around 0.5%.Equipped with an optimized active grid(AG),the OJF can reproduce controllable,fully-developed turbulent wind conditions with the turbulence intensity up to 24%,energy spectrum satisfying the five-thirds power law,and the uniform region close to 70%of the cross-sectional area of the test section.The turbulence intensity,integral length scale,Kol-mogorov length scale,and mean energy dissipation rate of the generated flow can be adjusted by varying the area of the triangular through-hole in the wings of the AG.展开更多
基金supported by the National Natural Science Foundation of China(Grant Nos.52306126,22350710788,12432010,11988102,92270203)the Xplore Prize.
文摘Configuring computational fluid dynamics(CFD)simulations typically demands extensive domain expertise,limiting broader access.Although large language models(LLMs)have advanced scientific computing,their use in automating CFD workflows is underdeveloped.We introduce a novel approach centered on domain-specific LLM adaptation.By fine-tuning Qwen2.5-7B-Instruct on NL2FOAM,our custom dataset of 28,716 natural language-to-OpenFOAM configuration pairs with chain-of-thought(CoT)annotations enables direct translation from natural language descriptions to executable CFD setups.A multi-agent system orchestrates the process,autonomously verifying inputs,generating configurations,running simulations,and correcting errors.Evaluation on a benchmark of 21 diverse flow cases demonstrates state-of-the-art performance,achieving 88.7%solution accuracy and 82.6%first-attempt success rate.This significantly outperforms larger general-purpose models such as Qwen2.5-72B-Instruct,DeepSeek-R1,and Llama3.3-70B-Instruct,while also requiring fewer correction iterations and maintaining high computational efficiency.The results highlight the critical role of domain-specific adaptation in deploying LLM assistants for complex engineering workflows.Our code and fine-tuned model have been deposited at https://github.com/YYgroup/AutoCFD.
基金supported by the National Natural Science Foundation of China(Grant Nos.91841302,11925201,and 11988102)the National Key Research and Development.Program of China(Grant No.2020YFE0204200)。
文摘We develop and assess a model of the turbulent burning velocity ST over a wide range of conditions.The aim is to obtain an explicit ST model for turbulent combustion modeling and flame analysis.The model consists of sub models of the stretch factor and the turbulent flame area.The stretch factor characterizes the flame response of turbulence stretch and incorporates detailed chemistry and transport effects with a lookup table of laminar counterflow flames.The flame area model captures the area growth based on Lagrangian statistics of propagating surfaces and considers the effects of turbulence length scales and fuel characteristics.The present model predicts sT via an algebraic expression without free parameters.We assess the model using 490 cases of the direct numerical simulation or experiment reported from various research groups on planar and Bunsen flames over a wide range of conditions,covering fuels from hydrogen to n-dodecane,pressures from 1 to 30 atm,lean and rich mixtures,turbulence intensity ratios from 0.1 to 177.6,and turbulence length ratios from 0.5 to 66.7.Despite the scattering sT data in the literature,the comprehensive comparison shows that the proposed ST model has an overall good agreement over the wide range of conditions,with the averaged modeling error of 28.1%.
基金sponsored by King Abdullah University of Science and Technology(KAUST)the National Natural Science Foundation of China(Grant No.91841302)。
文摘We propose a new flame index for the transported probability density function(PDF) method. The flame index uses mixing flux projections of Lagrangian particles on mixture fraction and progress variable directions as the metrics to identify the combustion mode, with the Burke-Schumann solution as a reference. A priori validation of the flame index is conducted with a series of constructed turbulent partially premixed reactors. It indicates that the proposed flame index is able to identify the combustion mode based on the subgrid mixing information. The flame index is then applied the large eddy simulation/PDF datasets of turbulent partially premixed jet flames. Results show that the flame index separate different combustion modes and extinction correctly. The proposed flame index provides a promising tool to analyze and model the partially premixed flames adaptively.
文摘Quantum computing has grown substantially over the past four decades,but whether it can outperform classical methods in practical use remains uncertain[1].Fluid dynamics simulation,challenging in classical physics but vital for applications,is a potential area for showcasing quantum advantage.The quantum computing for fluid dynamics(QCFD)[2]is expected to efficiently simulate intricate turbulent flows with high Reynolds numbers.This capability is crucial for critical applications,including aircraft design and weather forecast.
基金This work has been supported in part by the National Natural Science Foundation of China(Grant Nos.11925201 and 11988102)the National Key R&D Program of China(Grant No.2020YFE0204200)the Xplorer Prize.
文摘Bio-inspired micro-air-vehicles(MAVs)usually operate in the atmospheric boundary layer at a low Reynolds number and complex wind conditions including large-scale turbulence,strong shear,and gusts.We develop an open jet facility(OJF)to meet the requirements of MAV flight experiments at very low speed and high turbulence intensity.Powered by a stage-driven fan,the OJF is capable of generating wind speeds covering 0.1–16.8 m/s,with a velocity ratio of 100:1.The contraction section of the OJF is designed using an adjoint-driven optimization method,resulting in a con-traction ratio of 3:1 and a length-to-diameter ratio of 0.75.A modularized design of the jet nozzle can produce laminar or high-turbulence wind conditions.Flow field calibration results demonstrate that the OJF is capable of producing a high-quality baseline flow with steady airspeed as low as 0.1 m/s,uniform region around 80%of the cross-sectional test area,and turbulence intensity around 0.5%.Equipped with an optimized active grid(AG),the OJF can reproduce controllable,fully-developed turbulent wind conditions with the turbulence intensity up to 24%,energy spectrum satisfying the five-thirds power law,and the uniform region close to 70%of the cross-sectional area of the test section.The turbulence intensity,integral length scale,Kol-mogorov length scale,and mean energy dissipation rate of the generated flow can be adjusted by varying the area of the triangular through-hole in the wings of the AG.
