Exploiting non-equilibrium turbulence in design using DNS and Machine Learning

点击翻译按钮获取中文摘要

The current trend in fan design for large civil turbofans is to reduce the speed, and increase the diameter of the fan. This gives a reduced fan pressure ratio and increased bypass ratio, improving the propulsive efficiency of the engine, however these factors present a challenge to the fan designer, with new areas of the design space being explored, well away from the majority of current designs and experience. This makes accurate low order (Reynolds Averaged Navier Stokes - RANS) models crucial, however previous work has identified shortcomings with the prediction of the non-equilibrium region (where turbulence production exceeds dissipation) of turbulent boundary layers by an industry standard turbulence model in subsonic compressor cases.This project aims to gain an understanding of loss mechanisms in the non-equilibrium region of transonic turbulent boundary layers, and investigate how low order models may be improved to better capture these phenomena by answering the following research questions:1. What is the importance of non-equilibrium turbulence for transonic boundary layers? 2. Do current industry standard turbulence models capture non-equilibrium effects? 3. If current models are inadequate, how may they be modified to improve the prediction of non-equilibrium behaviour? The project will utilise high fidelity computational fluid dynamics techniques that resolve all (Direct Numerical Simulation - DNS) or much (Large Eddy Simulation - LES) of the range of scales of turbulent structures in the flow, giving a detailed insight into the flow physics that is not available with conventional (RANS) CFD or experiments. With the data generated with this work, innovative data-driven techniques in turbulence modelling will be explored to improve the prediction of non-equilibrium phenomena in low order methods suitable for use in industry for design. Efforts will also be made to improve existing turbulence models, such as by tuning modelling constants with reference to the DNS data to better predict these flows.