Development of liquid-sprays numerical modelling approaches for low-emission gas turbine combustion systems

Computational Fluid Dynamics (CFD) is used to guide the design of novel low-emission gas turbine combustion technologies. For combustion systems using liquid fuels, injecting the spray in crossflow, coupled with enhanced turbulence can achieve superior mixing characteristics and reduce emissions. The development of advanced mixing technologies requires accurate predictions of the spray characteristics. However, atomization is often modelled using the Lagrangian approach with deterministic breakup models instead of the higher-fidelity interface-capturing methods, such as Volume-of-Fluid (VOF), due to their high computational cost. The breakup models used in state-of-the-art CFD and developed specifically for liquid jets in crossflow (LJIC) rely on empirical constants and/or correlations that need to be calibrated with experimental data, which is not practical. The practicability of modelling LJIC could be improved by using a stochastic secondary droplet (SSD) breakup model that does not require exhaustive calibration. A better compromise between computational cost and accuracy could also be achieved by coupling the Lagrangian and VOF approaches. However, very few studies have assessed or validated their predictive capabilities for LJIC, especially for fuels and under more representative gas turbine conditions. This research aimed to improve the practicability of modelling turbulent liquid fuel jets in crossflow (LFJIC) and to contribute to the knowledge of the Lagrangian and hybrid approach through new assessment and validation work. Lower and higher-fidelity modelling approaches for turbulent LFJIC were assessed and validated using experimental data over a wide range of pressures, momentum flux ratios and Weber numbers. This comprised an investigation of the effects of different sub-models and boundary conditions on the flow properties, predictive capabilities, and computational cost. This collaborative research between Cranfield University and Siemens Energy has led to the development and validation of novel stochastic numerical methodologies as well as a better understanding of LFJIC modelling. The hybrid resolved method proposed offers superior predictive capabilities for the overall droplet velocities and droplet diameters in the leeward region but is on average 17 times more computationally expensive than the URANS Lagrangian methodology. The optimized stripping method proposed reduces the cost of the hybrid approach by 78% and the average relative difference between their predictions is less than 15%. Thus, their trade-offs between accuracy and computational cost were deemed reasonable and guidelines were formulated to help determine the most suitable methodology for future research work on turbulent LFJIC. The research outcomes are already being used to hasten the design, development, and delivery of novel low-emission gas turbine combustion systems.