Session: 04-08 Combustion Modeling I

Paper Number: 151529

Large Eddy Simulation of the PRECCINSTA Combustor Using a Native GPU Solver 

The recent development of GPU hardware enables the adaptation of large-scale unsteady CFD simulations across various applications and industries. The focus lies in the Large Eddy Simulation (LES), where all the relevant turbulence scales are resolved in time and space. This necessitates the use of fine meshes, long-time transient computations, and complex physics.

One of the first industrial sectors to adopt LES is Gas Turbine Combustion (GTC). For such applications, simplified turbulence models such as RANS are not reliable due to complex mixing processes, which require the resolution of the interaction between turbulence eddies and combustion processes. While LES has already been utilized in Ansys Fluent^® for many years, such simulations are often confined to relatively low resolution and may result in run-times which are not convenient within combustor design cycles. Therefore, the availability of the newly developed, Native Ansys CFD^® GPU solver may offer the potential to expand the application of LES to larger meshes and more complex physics/chemistries used by the GTC application, which can substantially reduce the computational time while retaining sufficient solution accuracy.

The latest Native GPU solver built within Ansys Fluent^® 2025R1 has been thoroughly tested on various reacting flow cases, and consistent solutions have been achieved between the GPU and CPU solvers. In the meantime, the best practice numerical, physical and chemistry models in the CPU solver are continuously being implemented and validated in the GPU solver, enhancing the current GPU capabilities. For example, in this work the well-studied bluff-body stabilized methane/air combustor, PRECCINSTA burner, has been studied for GPU validation purposes.

For the GPU simulation of the present combustor, the LES method with WALE sub-grid scale model has been applied for turbulence, and the species transport-based, Finite Rate (FR) combustion model has been selected, utilizing the Stiff Chemistry solver.

The Rapid Octree meshing technique within Ansys Fluent^® has been utilized to generate three levels of mesh grids for mesh sensitivity analyses. The coarsest mesh has been found sufficient to achieve solution independence. Different numerical settings are then compared for the GPU solver, determining the most accurate and efficient numerical set for GPU simulations. A series of chemical reaction mechanisms have also been compared on the solution accuracy and computational speed. Finally, the above computed GPU results have been compared with the corresponding CPU solutions on accuracy and speed, and also validated against available experimental data.

It has been found that the GPU solver (with the newly developed "Optimized LES Numerics") can offer significant improvement on the computational speed compared to CPU, and also reduce the number of sub-iterations per time step required for solution convergence. For the same mesh resolution, the Optimized LES Numerics setting can largely increase the amount of turbulent vortices in the computational domain, improving the solution accuracy compared to other conventional numerical settings.

Based on the GPU calculations of the PRECCINSTA combustor in this work, the Ansys Native GPU solver with the optimized LES numerical settings may be applied to more complicated industrial combustors, offering accurate and cost-effective GPU solutions to the Gas Turbine Combustion, and contributing to the deployment of GPU solver to the gas turbine industry.

Presenting Author: Yu Xia Ansys UK Ltd.

Presenting Author Biography: Yu Xia is a Senior R&D Engineer at Ansys UK Ltd. He achieved his PhD degree from Imperial College London in 2019, and has research interests on turbulence, combustion, multiphase flow and uses CFD to simulate various academic and industrial cases. He received the ASME Turbo Expo Early Career Engineer Travel Award in 2024, the ASME Turbo Expo Young Engineer Travel Award in 2018, and has published 8 top journal articles including Combustion and Flame, and over 30 international conference papers. More details can be found on my personal website: https://sites.google.com/view/steven-xia/

Authors:

Yu Xia Ansys UK Ltd.
Ashwini Dalvi Ansys Software Pvt. Ltd.
Florian Menter Ansys Germany GmbH
David Flad Ansys Germany GmbH
Andreas Hueppe Ansys Germany GmbH