Turbulent mixing of different temperature fluids in T-junction geometries is a technically critical issue for the safe operation of power plants. Due to the strong flow deformation, the scale separation assumption is not respected locally, limiting the applicability of classic unsteady Reynolds-averaged Navier-Stokes (URANS) models, which are unable to deliver the required accuracy in the prediction of temperature fluctuations. On the contrary, eddy resolving methods, and in particular large eddy simulation (LES), can provide reliable results at a computational cost that is still impracticable for the industry.

A robust second-generation URANS (2G-URANS) model was recently proposed at MIT, which aims at locally resolving complex flow structures. In the present paper, the performance of the structure-based (STRUCT) model is assessed specifically against low Reynolds number (??????=4,485) DNS data on a T-junction case. Velocity and temperature distributions in the mixing region are compared between URANS, STRUCT and LES solutions and the reference DNS data. The STRUCT model demonstrates significant advancement in the ability to model the thermal striping phenomena. Its application produces accurate predictions of the flow behavior on coarse URANS computational grids, with a large cost saving in comparison to LES.