We undertake high-fidelity simulation and data-driven modelling with application to transport, power generation and energy systems. The focus is on better understanding and predicting turbulent flows to improve efficiency, reliability and reduce noise emissions.
- Investigation of losses in gas turbines
- Aerodynamic noise generation mechanisms and mitigation approaches
- Simulation and modelling of reacting flows
High-fidelity simulation of turbulent reacting flows
Dominic Man Ching Ma
Understanding the physics of turbulent reacting flows fundamental in many engineering applications, such energy generating gas turbines and automotive engines, is of key importance to designing cleaner and more efficient devices. High-fidelity simulations of turbulent reacting flows, which require consideration of both turbulence and chemical reactions are increasingly in demand to aid in understanding flame and turbulence dynamics, and in developing numerical models. This project involves conducting high-fidelity simulations of turbulent reacting flows with HiPSTAR. The data is used to examine flame and turbulence dynamics of interest, and in numerical model development.
[Image: Reaction rate of a turbulent premixed jet flame in hot coflow.]
Direct Numerical Simulations of Fluid Flow and Heat Transfer in a Annular Rotating Cavity
This project focuses on the numerical investigation of fluid flow and heat transfer in a rapidly rotating annular cavity. These annular cavities are an integral part of an internal cooling system of the high pressure compressor and heat transfer from these disks affects the clearances between the blades and the casing of the high pressure compressor, which is a crucial factor for the operational efficiency and safety of a gas turbine engine. The main objective of this project is to understand the flow structure inside these cavities which affects the heat transfer in this conjugate system and vice versa. Another aim of this study is to quantify the relationship between Nussel Number and Rayleigh number which characterizes the heat transfer efficiency.
[Image: Instantaneous non-dimensional temperature at mid-axial position of an annular closed cavity shows the formation of convection cells due to the centrifugal buoyancy induced flow.]
Data-driven turbulence modelling with application to turbomachinery
The prediction of turbulent flows is a nontrivial problem in all branches of engineering, in particular in gas turbines, where predictions of turbine blade temperature are crucial to engine efficiency or life. For example, the trailing edge of a high-pressure turbine blade, which represents the thinnest section in a blade geometry and is exposed to substantial heat load during operation, requires efficient cooling strategies. Given high computational cost of DNS/LES in most practical cases, RANS is still widely used as a means of modelling for designers. However, RANS poorly predicts the wall temperature. The aim of this project is therefore to develop non-linear Reynolds stress and heat flux closures with increased prediction accuracy for trailing-edge cooling slot cases through applying a form of supervised machine learning framework to high fidelity datasets. The constructed models can be easily implemented into a flow solver since the algorithm used is a form of symbolic regression known as “Gene Expression Programming”, such that the result of the optimisation is a tangible equation. The outcome of this project is, hence, intended to substantially reduce the cost of accurately predicting the blade temperature in turbomachinery applications for designers.
[Image: Instantaneous LES temperature field of a wall jet with finite slot lip thickness, representing the trailing edge of a high-pressure turbine blade, which is exposed to substantial heat load during operation and therefore requires efficient cooling strategies.]
Aerofoil trailing edge noise high-fidelity simulation
Trailing edge noise accounts for a significant part of aerofoil noise, and its prediction remains a difficult task: current aeroacoustics models often fail to give an accurate prediction of both the noise level and frequency spectrum.
This project aims to perform the first Direct Numerical Simulation (DNS) of aerofoil noise at practical conditions. This objective is made affordable by taking advantage of cutting-edge numerical technologies such as GPU accelerated computing.
High-fidelity results can then be used as an input to feed machine-learning tools, in order to develop new models for trailing edge noise that combine low computationnal cost and high accuracy.
[Image: Acoustic waves generated by the wake of an aerofoil (obtained with DNS).]
High-Fidelity Simulation of Gas Turbines for Model Development using Machine Learning
High-fidelity simulation of Gas Turbines at engine relevant Reynolds number is conducted, using the in-house compressible flow solver HiPSTAR. The research generates a gold-standard database covering different Reynolds numbers, inlet turbulence intensities, length scales, and exit Mach numbers, which is applied to investigate prevalent turbomachinery flow phenomena such as transition, transonic behaviour on blade suction-side and secondary flow caused by end-walls. Furthermore, this data set is also utilized for model development using machine learning tools developed by our group, which have been shown to improve the predictive accuracy of low-order model.
To find out more about Dr Zhao's work watch the video he produced for the APS conference.
[Image: Vortical structures in LES of high-pressure turbine end-wall simulation.]
Reduced-order modeling of membrane wings at low Reynolds numbers
Highly deformable wings are a characteristic of many small-scale natural flyers and they have recently found application in the design of Micro-Air Vehicles (MAVs). Small-scale flapping-flight aerodynamics for membrane wings is characterized by complex unsteady nonlinear fluid-structure interaction phenomena and it is a multi-disciplinary topic that represents a major challenge for scientists and engineers. The aim of the present project is to develop reliable low-order models for the prediction of the fluid-structure interaction of small-scale membrane wings that can help in the design of MAVs. The modeling approach combines a linear reduced-order framework for the unsteady aerodynamics based on Direct Numerical Simulation with a one-dimensional membrane equation. DNS is also used as a validation tool.
[Image: Vorticity field obtained from Direct Numerical Simulation of a pitching membrane wing for a chord-based Reynolds number of 1000.]
Turbulent Wakes: Investigation Prediction
Wake dynamics resulting from vortex shedding are incredibly complicated and prediction of such wakes with RANS modeling is a challenge today. High fidelity (HiFi) simulations of these wakes can offer insight into their evolution, which can be used to improve our understanding of the dynamics as well as existing RANS models. The investigation considers effect of streamwise pressure gradients and their impact on the overall wake evolution. The prediction looks at improving RANS models by extracting information from the HiFi data, through use of Machine Learning, to develop better turbulence closures
[Image: Contours of vorticity for a DNS of a flat plate wake at Re = 2000 based on plate height and freestream velocity.]
Aerofoil Trailing-Edge Noise Prediction using Low-Order Models and Machine Learning
We are developing a fast and accurate prediction-tool of the noise emanating from the flow of air over an aerofoil. Current tools for use in industrial design processes are anchored in empiricism. Our physics-driven approach is rooted in developing a better understanding of the turbulent flow field over the aerofoil surface. New high-fidelity data, from this group, will serve as input to the low-order models and artificial intelligence software.
Applications range from the acoustics of hand-dryers to those of wind turbines and airplane wings. Indeed, aerofoil noise is a limiting factor behind global trade and people transport, and the deployment of renewable energy generators.
[Image: Flow past an aerofoil - a 2D illustration of velocity streamlines and vorticity in the wake.]