Wednesday, November 11, 2015: 12:30 PM
150A/B (Salt Palace Convention Center)
Introducing a trace amount of polymer into the liquid turbulent flows can result in substantial reduction of friction drag. This rheological drag-reducing phenomenon has been widely used in fluid transport, such as the Alaska crude oil pipeline. However, the mechanism or the theoretical picture behind the phenomenon is not well understood. In order to gain more insight into this issue, we conduct direct numerical simulations (DNS) of fully-developed Newtonian and viscoelastic turbulent flows in large domains that imitate experiment length scales. In large-domain simulations, the flow shows different characteristics in different regions. In some areas hibernating turbulence is occurring, while in other areas the turbulent motions are much more active. To identify these characteristic regions, we apply a statistical method capable of partitioning a data set by its own algorithm based on a certain criterion. K-means clustering partitions the observations into k clusters by assigning each observation to its nearest mean called centroid. The resulting partition maximizes the between-cluster variance. In large-domain simulations, the observations are the flow variables from mesh points at the walls. Regions with different levels of drag are automatically identified by the partitioning algorithm. We find that the velocity profiles of the centroids exhibit characteristics similar to the individual coherent structures observed in minimal domain simulations. In addition, we observe that as viscoelasticity increases, polymer stretch becomes strongly correlated with wall shear stress. How viscoelasticity modifies the spatial-temporal dynamics in large-domain simulations is to be determined in this work.