An Assessment of Large-Eddy Simulation Toward Thermal Fatigue Prediction

A potential cause of thermal fatigue failures in energy cooling systems is identified with cyclic stresses imposed on a piping system. These are generated from temperature changes in regions where a cold flow is intensively mixed with a hot flow. Typical locations of such thermal mixing are T-junctions in nuclear reactor cooling systems. Turbulent mixing in a T-junction is investigated here using large-eddy simulation (LES). In general, LES can capture very well the mixing phenomena and the accompanying turbulent flow fluctuations that occur in a T-junction. Through a direct comparison with experimental results, an assessment of the accuracy of LES predictions is made for the Smagorinsky and the Vreman models. It is shown that the results obtained with the Vreman model are closest to the experimental results. The Smagorinsky model is found to provide the least accurate results. This is particularly detected in the near-wall regions that are of great importance in thermal fatigue predictions. Detailed numerical validation was performed with simulations using five different spatial mesh resolutions. These simulations show that computational meshes must resolve important turbulence length scales in order to obtain sufficiently accurate results. This accuracy assessment and error quantification are based on the integral and Taylor length scales of turbulence. For the investigated cases, the mesh resolution with average cell sizes of the order of λ/3 (three times smaller than the Taylor microscale length) is sufficient to give very similar results to those obtained on much finer meshes. An engineering estimation of the minimal mesh resolution gives an initial guideline for construction of computational meshes that allow for accurate predictions of turbulent mixing in a T-junction using LES. Additionally, analysis of the temperature measurement data at specified probe locations is presented along with a quantification of an error introduced by the applied LES method.