Search

The objective of this work is to compare a high-order solver with a low-order solver for performing large-eddy simulations (LES) of a compressible mixing layer. The high-order method is the Wave-Resolving LES (WRLES) solver employing a Dispersion Relation Preserving (DRP) scheme. The low-order solver is the Wind-US code, which employs the second-order Roe Physical scheme. Both solvers are used to perform LES of the turbulent mixing between two supersonic streams at a convective Mach number of 0.46. The high-order and low-order methods are evaluated at two different levels of grid resolution. For a fine grid resolution, the low-order method produces a very similar solution to the high-order method. At this fine resolution the effects of numerical scheme, subgrid scale modeling, and filtering were found to be negligible. Both methods predict turbulent stresses that are in reasonable agreement with experimental data. However, when the grid resolution is coarsened, the difference between the two solvers becomes apparent. The low-order method deviates from experimental results when the resolution is no longer adequate. The high-order DRP solution shows minimal grid dependence. The effects of subgrid scale modeling and spatial filtering were found to be negligible at both resolutions. For the high-order solver on the fine mesh, a parametric study of the spanwise width was conducted to determine its effect on solution accuracy. An insufficient spanwise width was found to impose an artificial spanwise mode and limit the resolved spanwise modes. We estimate that the spanwise depth needs to be 2.5 times larger than the largest coherent structures to capture the largest spanwise mode and accurately predict turbulent mixing.

Separated flow over a bluff body is analyzed via large eddy simulations. The turbulent flow around a square cylinder features a variety of complex flow phenomena such as highly unsteady vortical structures, reverse flow in the near wall region, and wake turbulence. The formation of spanwise vortices is often times artificially suppressed in computations by either insufficient depth or a coarse spanwise resolution. As the resolution is refined and the domain extended, the artificial turbulent energy exchange between spanwise and streamwise turbulence is eliminated within the wake region. A parametric study is performed highlighting the effects of spanwise vortices where the spanwise computational domain's resolution and depth are varied. For Re=22,000, the mean and turbulent statistics computed from the numerical large eddy simulations (NLES) are in good agreement with experimental data. Von-Karman shedding is observed in the wake of the cylinder. Mesh independence is illustrated by comparing a mesh resolution of 2 million to 16 million. Sensitivities to time stepping were minimized and sampling frequency sensitivities were nonpresent. While increasing the spanwise depth and resolution can be costly, this practice was found to be necessary to eliminating the artificial turbulent energy exchange.

Axisymmetric subsonic diffuser flows were calculated with the NPARC Navier-Stokes code in order to determine the effects various code features have on the flow solutions. The code features examined in this work were turbulence models and boundary conditions. Four turbulence models available in NPARC were used: the Baldwin-Lomax algebraic model, the Baldwin-Barth one-equation model, and the Chien kappa-epsilon and Wilcox kappa-omega two-equation models. The three boundary conditions examined were the free boundary, the mass flux boundary and the subsonic outflow with variable static pressure. In addition to boundary condition type, the geometry downstream of the diffuser was varied to see if upstream influences were present. The NPARC results are compared with experimental data and recommendations are given for using NPARC to compute similar flows.

Flows through a transonic diffuser were investigated with the PARC code using five turbulence models to determine the effects of turbulence model selection on flow prediction. Three of the turbulence models were algebraic models: Thomas (the standard algebraic turbulence model in PARC), Baldwin-Lomax, and Modified Mixing Length-Thomas (MMLT). The other two models were the low Reynolds number k-epsilon models of Chien and Speziale. Three diffuser flows, referred to as the no-shock, weak-shock, and strong-shock cases, were calculated with each model to conduct the evaluation. Pressure distributions, velocity profiles, locations of shocks, and maximum Mach numbers in the duct were the flow quantities compared. Overall, the Chien k-epsilon model was the most accurate of the five models when considering results obtained for all three cases. However, the MMLT model provided solutions as accurate as the Chien model for the no-shock and the weak-shock cases, at a substantially lower computational cost (measured in CPU time required to obtain converged solutions). The strong shock flow, which included a region of shock-induced flow separation, was only predicted well by the two k-epsilon models.

A comparative study was conducted for computational fluid dynamic solutions to flow over a backward-facing step. This flow is a benchmark problem, with a simple geometry, but involves complicated flow physics such as free shear layers, reattaching flow, recirculation, and high turbulence intensities. Three Reynolds-averaged Navier-Stokes flow solvers with k-epsilon turbulence models were used, each using a different solution algorithm: finite difference, finite element, and hybrid finite element - finite difference. Comparisons were made with existing experimental data. Results showed that velocity profiles and reattachment lengths were predicted reasonably well by all three methods, while the skin friction coefficients were more difficult to predict accurately. It was noted that, in general, selecting an appropriate solver for each problem to be considered is important.

A simple turbulence model that will be applicable to propulsion flows having both wall bounded and unbounded regions was developed and installed within the PARC Navier-Stokes code by linking two existing algebraic turbulence models. The first is the Modified Mixing Length (MML) model which is optimized for wall bounded flows. The second is the Thomas model, the standard algebraic turbulence model in PARC which has been used to calculate both bounded and unbounded turbulent flows but was optimized for the latter. This paper discusses both models and the method employed to link them into one model (referred to as the MMLT model). The PARC code with the MMLT model was applied to two dimensional turbulent flows over a flat plate and over a backward facing step to validate and optimize the model and to compare its predictions to those obtained with the three turbulence models already available in PARC.