High-fidelity simulation of a turbulent inclined jet in a crossflow

Film cooling systems induce a range of complex phenomena associated with the interaction of the coolant jet discharging into a crossflow. The jet-in-crossflow configuration has been extensively studied through experiments and direct/large-eddy numerical simulations. However, most previous simulations consider a laminar crossflow and a transverse jet originating from a fully developed pipe flow (Muppidi & Mahesh 2007). For numerical calculations to capture the physical phenomena relevant to film cooling applications (Acharya et al. 2001), it is necessary to include the effects induced by: (i) the recirculation occurring at the inlet of the short film cooling hole and (ii) the turbulent state of the main flow boundary layer. The first point requires including the plenum feeding the hole in the computational domain. The second condition can be fulfilled by imposing a realistic turbulent flow at the inflow of the developing crossflow. Most previous numerical studies could not afford to satisfy both requirements at the same time (Muldoon & Acharya 2009; Ziefle & Kleiser 2008) because of the prohibitive computational cost. In this work a film cooling flow replicating a parallel experimental study by Coletti et al. (2013) is investigated by means of Large Eddy Simulation (LES). A plenum with a quasi-quiescent flow is connected to a square duct through a short inclined pipe, injecting a non-buoyant contaminant into the duct flow. Both jet and crossflow are fully turbulent. The measurements are obtained by MRI-based techniques (Benson et al. 2010; Elkins et al. 2003) and include the three-dimensional velocity and concentration fields in the whole physical domain, along with Reynolds stresses measured by PIV along selected planes. The comprehensive experimental database allows a complete validation of the time-averaged calculated fields, as well as testing the sensitivity of the results to the turbulent Schmidt number employed in the subgrid scale model. The simulation, besides achieving higher spatial resolution than the experiments, provides additional information on the turbulent scalar fluxes and on the time-dependent nature of the momentum and scalar transport. The availability of the turbulent scalar fluxes allows direct evaluation of the turbulent diffusivity and turbulent Schmidt number, which are key parameters in commonly employed RANS solvers. The latter typically use isotropic formulations for the turbulent diffusivity, and in fact they perform poorly, especially close to injection (Kohli & Bogard 2005). The present calculation allows a priori evaluation of the scalar fluxes models. Furthermore, deviation from the Reynolds analogy and the gradient-diffusion hypothesis can be examined.