 |
Center for Nonlinear Studies |
Turbulent combustion is a highly non-linear phenomenon that involves many competing physical processes — such as chemical kinetics, turbulence, multi-phase transport and radiation — which all take place over a wide range of spatial and temporal scales. Since small-scale physics directly impact larger-scale behaviour, accurate mathematical representations of turbulent flames are extremely difficult to obtain because all of the physical processes and scales must be properly captured. This hinders our ability to fully understand turbulent flames since there are no experimental and numerical techniques which can fully capture all of the governing processes. High-fidelity numerical methods and high-performance computing (HPC) are playing an increasingly more significant part in understanding turbulent combustion. But there is currently a limit on what can be simulated because of the complexities in the physics and the range of scales that must be resolved. Therefore, mathematical models of these complex devices must rely heavily on engineering approximations and sophisticated numerical methods to represent the underlying physics and ensure that computations remain tractable. Both adaptive mesh refinement (AMR) and high-order spatial discretizations are effective methods for providing accurate solutions with minimal computational resources. AMR automatically adapts the computational grid to the numerical solution in order to treat problems with disparate length scales using a reduced number of mesh points. High-order discretization techniques offer the potential to significantly reduce the computational costs necessary to obtain accurate predictions when compared to lower-order methods. However, efficient, universally-applicable, high-order discretizations remain somewhat illusive, especially for more arbitrary unstructured meshes and for incompressible/low-speed flows. New, efficient, highly-scalable solution algorithms are also required so that simulations of complex practical flames remain tractable. To obtain solutions to such problems, new algorithms are required that can fully take advantage of the growing trend towards exascale computing with large, multi-core, multi-threaded, heterogeneous architectures. A novel, parallel high-order finite-volume scheme and block-based AMR algorithm for large-eddy simulation (LES) of reacting flows on unstructured mesh will be discussed.
Host: Mikhail Shashkov, XCP-4 Methods and Algorithms, 667-4400