Large Eddy Simulation
Most of the research currently taking place in the field of CFD concerns the study of turbulent flows. Almost any naturally occurring flow is turbulent, and hence it is important to be able to model turbulent flows accurately. To that end, many models have been put forth to provide accurate solutions to these flows. Large Eddy Simulation is simply one of these models.
The three major types of turbulence methodologies are Direct Numerical Simulation (DNS), Large Eddy Simulation (LES) and k-epsilon modeling. K-epsilon simply attempts to model the turbulence by performing time or space averaging. Under certain conditions this method can be very accurate, but it is not suitable for transient flows, because the averaging process wipes out most of the important characteristics of a time-dependent solution. Direct Numerical Simulation, on the other hand, attempts to solve all time and spatial scales. As a result, the solution is very accurate. Unfortunately, DNS is unrealistic for 99.9% of CFD problems because it is computationally unrealistic. That is, to resolve all spatial and temporal scales, the spatial and temporal grids would need to be extremely small, resulting in a problem which would take an extraordinarily long time to solve with today's technology.
One compromise between these two methods is Large Eddy Simulation. This technique was originally implemented in the 1970s by atmospheric scientists to study the weather. Since that time it has been utilized in almost every engineering field. Large Eddy Simulation seeks to directly solve large spatial scales (like DNS), while modeling the smaller scales (k-epsilon). The basis for this is two-part. First, the larger scales carry the majority of the energy, and hence are more important. Second, the smaller scales have been found to be more universal, and hence are more easily modeled. The resulting methodology is a hybrid between these two methods, which involves the filtering of the Navier-Stokes equations to separate those scales which will be modeled from those which will be solved for directly. This method is useful for the CFD research group here at Texas A&M because it allows us to generate useful solutions to transient flows, while still maintaining computationally realistic problems.