In recent years, high performance computing-based computational fluid dynamics has been able to tackle an increasing number of real world engineering problems. These techniques, namely large eddy simulation (LES) and direct numerical simulation (DNS), require very minimal empirical modeling and rely largely on first principles to resolve the flow field. While largely relegated to academia in the past, as supercomputing progresses, such approaches will become accessible for a much larger portion of industrial flows.
 
The potential increased accuracy and reduction in uncertainty will definitely benefit some particular applications (i.e., natural convection), but it should not be understood as the main potential of these approaches. In fact, they represent an invaluable window in the flow physics, especially when used in conjunction with advanced methods such as linear stability analysis, proper orthogonal decomposition, wavelet analysis, quadrant analysis, etc. The knowledge obtained through these "numerical experiments" complements the irreplaceable knowledge gained by the now much more expensive physical experiments.
 
In this talk we will examine how these methods have helped us gain a much deeper understanding of the flow physics in a variety of geometries, with a particular focus on the flow in fuel assemblies.