GPU Accelerated Large Eddy Simulations of Combustion Instabilities in Industrial Gas Turbines Under EGR Conditions

Abstract Exhaust gas recirculation (EGR) is a promising technology for increasing CO2 concentrations in gas turbine exhaust, thereby enhancing the efficiency of downstream carbon capture processes. With EGR, part of the engine exhaust gas is recirculated back to the combustor inlet, mixed with fresh air, and reintroduced into the combustion system. High recirculation rates are critical to maximizing the overall carbon capture potential; however, too much recirculation can induce combustion instabilities and present other operational challenges. In this study, large eddy simulations (LES) are used to investigate the effects of EGR on combustion instabilities and emissions in Solar Turbines’ Taurus 60 SoLoNOx combustor. Grid-converged simulations are first performed for a double-injector combustor geometry using the Cadence Fidelity LES solver and a flamelet progress variable (FPV) turbulent combustion model. The LES model is validated for an atmospheric test-rig condition, showing excellent agreement with the experimental temperature profile at the combustor exit. The validated LES model is then applied to high-pressure full-load conditions for a wide range of EGR levels to quantify the impact of EGR on combustion instability and NOx emissions. The sensitivities of instability and NOx predictions to the choice of chemical reaction mechanism, as well as the inclusion of the compressor exit guide vane and first-stage turbine nozzles, are examined in detail. Overall, as EGR levels increase, instabilities start to grow exponentially once the EGR level reaches 80% of the maximum value tested, consistent between LES and the experiments. The choice of reaction mechanism barely affects the predictions of combustion instability while USC-Mech II shows a more accurate NOx prediction compared to the UCSD mechanism. The inclusion of geometric details upstream and downstream of the computational domain, in contrast, is shown to significantly enhance the predicted instability growth with increasing EGR. Lastly, an LES of the T60 combustor using the full 360° geometry is performed for the highest EGR condition. Compared to the double-injector simulation, more pressure fluctuation modes are predicted while the impact on NOx prediction is minimal. Moreover, the GPU-resident solver demonstrates excellent scalability and efficiency for the full-scale annular simulations (83 million mesh), providing two orders of magnitude speedup over conventional CPU-based simulations.