(165d) GPU Accelerated, High Fidelity Flamelet Modelling for Industrial Combustion Applications

The power generation and process industry accounts for over one-third of the global total primary energy demand. This demand has traditionally been met with combustion of fossil fuels. There are ever more challenges in the design of industrial combustion systems to ensure that emissions are reduced. This can include specific consideration in the design process as well as consideration of alternative fuels such as hydrogen and ammonia.

Understanding the behaviour of combustion systems early in the design process can allow for improved designs, however, it is not practical to carry out a full range of tests prior the execution of the system. CFD (Computational Fluid Dynamics) provides a methodology to interrogate a design space and understand underlying features during the design process. The more efficiently CFD simulations can be executed, the greater the exploration of the design space.

High-fidelity simulation such as Large Eddy Simulations (LES) can be used to assess combustor flow and mixing fields, flame position and flame dynamics and methods to reduce the computational costs of such simulations are desirable. One such method to achieve this is to utilize advanced High Performance Computing (HPC) architectures such as General Purpose Graphical Processing Units (GPGPUs).

In this paper, the Simcenter STAR-CCM+ ® GPGPU enabled solver is used. This includes efficient implementation of boundary conditions, complete discretization, linear solve and postprocessing all performed natively on the GPU without need for repeated CPU-GPU data migration,

With the ability to run LES and flamelet based combustion models using GPGPUs, is used to assess the benefits of using GPGPU architecture compared to traditional Central Processing Units (CPUs). LES simulations using the Flamelet Generated Manifold (FGM) combustion model are applied to a lifted flame and an industrial burner.

Numerical simulations using GPGPUs and CPUs are compared in terms of solution similarity, accuracy in relation to experimental data, computational cost and power consumption. The computational performance of the FGM model with GPGPU is quantified through comparison to a non-reacting simulation on the same computational grid.

It is demonstrated that the use of GPGPU architectures can result in high-fidelity combustion modelling within the required turn-around times of engineering design cycles whilst maintaining a similar solution and level of accuracy as a traditional CPU based simulation.