In the past 40 years, Computational Fluid Dynamics (CFD) has matured considerably, transitioning from a tool to compute forces on an airfoil to one that can simulate flows, temperatures and compositions inside chemical reactors. Today, CFD is widely accepted and frequently applied by furnace and burner manufacturers and users to improve efficiency and quality of heat-treating, melting, quenching and other high-temperature operations.
While CFD was once limited to “ivory tower” practitioners who could muster massive computational resources and tweak parameters to agree with well-controlled experiments, today there are a number of commercially available software packages that can be run on a standard desktop computer with reasonable accuracy and relatively short convergence times.
The “holy grail” of CFD is to generate an exact solution to the Navier-Stokes equations (mass, momentum, energy and species conservation) in three spatial dimensions with unsteady flows, radiative heat transfer and multistep chemistry. While this goal is being achieved today for simple flows and reactions using a computational technique called Direct Numerical Simulation (DNS), most combustion problems are still impractical to solve by DNS even with state-of-the-art computing resources.
Two other techniques are used today that employ simplifications to the full Navier-Stokes treatment. The more exact (and more computationally intensive) method is called Large Eddy Simulation (LES), and the more widely used desktop technique is called Reynolds Averaging Navier Stokes (RANS).
LES modeling is used where details associated with turbulent-flow features play a significant role in predicting the outcome of a computation. Prediction of NOx in nozzle-mixed burner combustion is often approximated more effectively with an LES model.
RANS modeling is often used when the details of the turbulent structures are less important to the outcome of the simulation. With RANS, the turbulence is typically described by a two-parameter model (k and ε), which satisfies average fluid constraints on energy and momentum without having to compute feature-by-feature velocity details of small turbulent vortices.
According to David Eby, managing engineer at Exponent and former manager of advanced methods at CD-Adapco, “RANS packages are often used to solve the flow and temperature fields first, and then a post-processor is applied to compute NOx concentrations. The newest tool that is coupled with some RANS solvers (e.g., STAR-CCM+) is called Digital Analysis of Reaction Systems (DARS). This technique is being successfully applied to a variety of combustion problems, including non-premixed, partially premixed and fully premixed combustion.”
A final word of caution to CFD users is embodied in the acronym GIGO (garbage in, garbage out). No matter how computationally powerful a software product may be, it can produce results that are horribly misleading if the user applies inappropriate boundary or initial conditions or chooses the wrong tool or wrong simplifications for the application. For a furnace user who would like to evaluate whether a particular burner can be tweaked to improve NOx emissions, rather than installing a new low-NOx burner, consultation with a CFD expert may prove invaluable.