Data-Driven Modeling of Turbulence and Heat Flux for Film Cooling Flow

• Applicant: Professor Heng Xiao, Ph.D.
• Subject Area: Fluid Mechanics
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 551388164

The joint research team between GIST and Uni-Stuttgart merges the knowledge of aviation aerodynamics and scientific machine learning to achieve the research goal. We aim to develop an innovative ML-based turbulence and heat-flux model that will significantly advance the current simulation capabilities to predict challenging turbulent thermal fluid flows for future aviation propulsion. In particular, we focus on predicting jet-in-crossflow in film cooling for turbines, which are rich physics and a critical flow phenomenon for efficient aero-engines. The research team merges the knowledge of aviation aerodynamics and scientific machine learning (ML) to achieve the research goal. We aim to develop an innovative ML-based turbulence and heat-flux model that will significantly advance the current simulation capabilities to predict challenging turbulent thermal fluid flows for future aviation propulsion. In particular, we focus on the prediction of jet-in-crossflow in film cooling for turbines, which as rich physics and a critical flow phenomenon for efficient aero-engines. The proposed research will: (1) Construct high-fidelity dataset for machine learning of turbulence and heat-flux models, (2) Develop data-driven, coupled turbulence and heat flux model with machine learning, (3) Integrate ML-based model in open-source industrial standard flow solvers, and (4) Extract modeling and design insight from the learned model and transfer know-how to relevant industries. To achieve these objectives, we will perform large eddy simulations to generate high-fidelity data for a family of film cooling flows under diverse inflow conditions, for which an efficient AI-based inflow generation technique will be developed. The benchmark dataset, including full-field data of turbulent stresses and heat fluxes and sparse measurements of cooling efficiency, will be used to train the neural network models. The ML-based model for coupled turbulence and heat flux will be designed to ensure frame-invariance. We will include a flow solver in the training process to ensure the robustness and generalizability of the learned model. We leverage gradient-based, multi-objective optimization to address different physical quantities in the data. The learned model will then undergo a systematic evaluation that includes extracting physical insights. Finally, the validated models will be integrated into widely used open-source flow solvers to enhance the accuracy of their predictive simulations for complex film cooling flows in industrial settings. The joint project will develop an innovative turbulence model that can accurately predict jet-in-crossflow phenomenon, particularly film cooling flow in gas turbine engines. The innovative turbulence model will (1) overcome the limitations of the conventional models, (2) open up new efficient design space for turbine cooling methods, and eventually (3) clean aviation technology with efficient aero-engines.

• DFG Programme: Research Grants
• International Connection: South Korea
• Cooperation Partner: Professor Solkeun Jee, Ph.D.