GPU-Native Implementation of the Thickened Flame Model (TFM) and Its Combination With Cell Agglomeration: Application to the Lean Premixed PRECCINSTA Burner

There is a growing interest in performing high-fidelity simulations of reacting flow in a computationally tractable manner. A popular approach towards achieving this objective is the use of GPUs, which can enable computationally efficient calculations. In the context of reacting flow simulations, there have only been a few efforts that have explored the use of turbulent combustion models that directly incorporate finite-rate chemistry effects. As part of the current work, we implement the thickened flame model (TFM) in the ANSYS FLUENT GPU-native solver. The GPU-native implementation is validated by performing large-eddy simulation/TFM of the stable variant of the lean premixed PRECCINSTA burner. The resolved statistics obtained from the GPU solver show good agreement with the measurements and the previously validated FLUENT CPU solver. A comparison of wall-clock time shows the advantages of using a GPU-native solver. Finally, we consider the combination of the GPU solver with a cell agglomeration algorithm, which is a technique for accelerating the reaction integration. The combined method provides further reduction in the wall-clock time per time step with minimal loss of accuracy.