Conservative Formulation of the k-ϵ Turbulence Model for Shock–Turbulence Interaction
Krishnendu Sinha∗ and S. J. Balasridhar† Indian Institute of Technology Bombay, Mumbai 400076, India
DOI: 10.2514/1.J052289 Reynolds-averaged turbulence models can result in large numerical error at flow discontinuities like shock waves. This is due to the nonconservative nature of the source terms in the governing equations. In this paper, the k-ϵ turbulence model is used to compute the canonical interaction of a normal shock with homogeneous isotropic turbulence. The characteristics of the nonconservative error is studied as a function of shock strength, grid resolution, and upstream turbulence level. The model equations are cast in an equivalent conservation form that gives physically consistent results at a shock wave. The predicted amplifications of turbulent kinetic energy and its dissipation rate match direct numerical simulation data and linear theory estimates over a range of Mach numbers.
Downloaded by UNIV OF TEXAS AUSTIN on June 10, 2013 | | DOI: 10.2514/1.J052289

0 b1 c0 ,

c1 , c2 f, g k M Mt Reλ u x δ ϵ κ0 λ μ μT ρ σk , σϵ Subscripts 0 1 2

= = = = = = = = = = = = = = = = =

shock-unsteadiness damping parameter k-ϵ model parameters new turbulence variables turbulent kinetic energy Mach number turbulent Mach number Reynolds number based on Taylor microscale fluid velocity shock-normal direction computed mean shock thickness turbulent dissipation rate wave number corresponding to most energetic scale Taylor microscale molecular viscosity turbulent eddy viscosity fluid density Prandtl numbers for k and ϵ

= = =

inlet station upstream of shock wave downstream of shock wave



NTERACTION of shock wave with turbulent boundary layers is common in many high-speed flows. Examples include deflected control surfaces, supersonic and hypersonic inlet ducts, multibody aerodynamics, and wing–body junctions. Presence of strong shock waves causes boundary-layer separation, which can generate additional shocks and expansion waves. Shear-layer reattachment downstream of the shock interaction often leads to localized high pressure and heat flux to the vehicle surface. A separation bubble inside an inlet duct acts as a blockage to the flow and can cause unstart. Accurate numerical prediction of shock–boundary-layer interaction (SBLI) flows is a challenging task, especially in the presence of strong shock waves. Several research efforts [1–4] have
Received 20 August 2012; revision received 21 December 2012; accepted for publication 2 January 2013; published online 23 April 2013. Copyright © 2013 by Krishnendu Sinha. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission. Copies of this paper may be made for personal or internal use, on condition that the copier pay the $10.00 per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923; include the code 1533-385X/13 and $10.00 in correspondence with the CCC. *Associate Professor, Department of Aerospace Engineering. Senior AIAA Member. † Graduate Student, Department of Aerospace Engineering.


been directed toward this goal. See [5,6] for a comprehensive survey of previous work. The interaction of turbulent fluctuations in the boundary layer with the shock wave lies at the heart of these phenomena. Shock– turbulence interaction has therefore been the focus of several studies, some of which are discussed next. Homogeneous isotropic turbulence passing through a normal shock is possibly the most fundamental shock–turbulence interaction. The mean flow is onedimensional and steady and is therefore uniform upstream and downstream of the shock wave. The jump in the mean flow quantities across the shock is governed by the Rankine–Hugoniot relations. Compared to shock–boundary-layer interaction, the model problem does not have additional complexity due to the flow separation, streamline curvature, and boundary-layer velocity gradients. Shock-homogeneous turbulence interaction has been extensively studied using direct numerical simulation (DNS) [7–10]. This canonical interaction is also amenable to theoretical analysis using rapid distortion theory [11,12] and linear interaction analysis [13,14]. Some limited experimental data are also available in the literature [15]. In spite of the geometrical simplicity, the model problem exhibits a range of physical effects, like generation of acoustic waves, baroclinic torques, and unsteady shock oscillations. Physical insight obtained in this canonical problem has proved useful in developing advanced turbulence models for shock–turbulence interaction [16–18]. Figure 1, reproduced from [18] with permission shows the variation of turbulent kinetic energy k as homogeneous isotropic turbulence interacts with a nominally normal shock wave. Here, x is the shock-normal direction, and the shock is located at x ˆ 3. The mean flow Mach number upstream of the shock is 1.5, the turbulent Mach number is 0.17, and the Reynolds number based on Taylor microscale is 6.7. The data correspond to the DNS of Jamme et al. [9], where the incoming turbulence field is primarily composed of vortical fluctuations. The magnitude of thermodynamic fluctuations is relatively small. The turbulent kinetic energy plotted in Fig. 1 is normalized by its value immediately upstream of the shock. A characteristic length of 2λ is used for normalizing the shock-normal distance, where λ is the Taylor microscale in the incoming turbulent flow. In the absence of mean velocity gradient upstream of the shock, the turbulent kinetic energy (TKE) decays from its inlet value. There is an increase in turbulence level across the shock, followed by further decay in the downstream flow. An amplification of the turbulent dissipation rate at the shock results in a faster decay of TKE behind the shock wave. Large values of turbulent kinetic energy are reported in the vicinity of the shock wave. These are artifacts of the unsteady shock oscillation, and they do not represent turbulent fluctuations [8]. Computation of the test case using Reynolds-averaged Navier– Stokes (RANS) equations is also presented in the figure. The RANS approach computes the mean flowfield and includes the effect of

AIAA Early Edition / 1

The turbulence models do not reproduce this transfer between the acoustic and vortical modes.2 3 AIAA Early Edition / SINHA AND BALASRIDHAR u ~ ρ Standard model ~ ∂k 2 ∂u 0† − ρ  k … 1 − b1 ϵ ˆ− ρ ∂x 3 ∂x (1) 2 DNS Realizable model 0 ˆb 1−M † represents shockwhere the parameter b1 | DOI: 10. [16] study the physical processes involved in shock/ homogeneous turbulence interaction using linearized governing equations. in the limit of vanishing turbulent dissipation at the shock. A detailed analysis and modeling of the ϵ amplification is presented in a recent paper by Sinha [18]. is evaluated against linear interaction analysis (LIA) results. brings down the postshock turbulence level. It is argued in [16] that the high level of turbulence production is caused by the breakdown of the eddy-viscosity assumption in the highly nonequilibrium region of a shock wave. . The ϵ equation form of the Reynolds stress u k 1 Shock-unsteadiness model 0 0 1 2 3 4 5 6 x Downloaded by UNIV OF TEXAS AUSTIN on June 10.3). reprinted with permission from [16]. It is. exhibit qualitative and quantitative disagreement with the linear theory.5 shock/homogeneous turbulence interaction using the shock-unsteadiness model shows good match with the postshock DNS data. 2. but its net effect on TKE amplification is included in the shock-unsteadiness model. therefore. Computation of the Mach 1. This is due to the decay of acoustic energy generated at the shock and its transfer to the kinetic form. overpredicts TKE for stronger shock waves (see Fig. on the other hand. the postshock k and ϵ levels increase dramatically (see Fig. TKE and ϵ amplifications obtained from DNS of Larsson and Lele [10] are found to match LIA far-field predictions. turbulent fluctuations using a turbulence model. Different versions of the k-ϵ model are compared with DNS data [9]. the turbulent viscosity is suppressed in the production term so as to minimize the error due to the limitations in the Boussinesq approximation. 10 5 Realizable model Linear analysis 8 Realizable model 6 4 k2 / k1 ε2 / ε1 3 µT = 0 Linear analysis Sinha et al. Conventional turbulence models like standard k-ϵ and k-ω models predict high amplification of turbulence across the shock (see Fig. 2013 | http://arc. Further. [16] 4 2 2 1 1 2 Shock-unsteadiness model 3 4 5 0 1 2 3 µT = 0 4 5 M M a) b) Fig. The standard k-ϵ model solution is found to be highly sensitive to the grid-point distribution at the shock. Setting μT ˆ 0 yields the isotropic g 0 02 ˆ 2k∕3 at the shock.J052289 Fig. the shock-unsteadiness model matches the theoretical predictions for the entire range of Mach numbers. 2 Amplification in turbulent kinetic energy and turbulent dissipation rate in shock/homogeneous turbulence interaction as a function of upstream mean flow Mach number. however. Compressibility corrections [21]. This can result in significant error in the pressure and surface heat-flux predictions [20]. and consequently a high turbulent viscosity downstream of the shock. The realizable k-ϵ model predictions are significantly higher than theory for all shock strengths. often leads to delayed flow separation in shock–turbulent boundarylayer interactions [19]. The realizable and μT ˆ 0 models.aiaa. do not improve turbulence levels significantly [16]. The k-ϵ model with μT ˆ 0 matches LIA results for weak shocks (M < 1.∞ …1 − e unsteadiness effect. 2b. Also. Similar results for the amplification of turbulent dissipation rate ϵ are shown in Fig. As the grid is refined to get a thinner shock wave.5. The theory assumes linear interaction of incoming turbulent fluctuations with the shock.∞ ˆ 0. Viscous effects are neglected to compute the jump in turbulence quantities across the shock wave. A comparison of turbulence amplification for varying shock strength is shown in Fig. We note that there is a nonmonotonic variation in turbulent kinetic energy immediately downstream of the shock wave (3 < x < 4). Overprediction of postshock turbulent kinetic energy.2514/1. but the predictions are still appreciably higher than DNS data (see Fig. and b1. 2. By comparison. which is modeled as a discontinuity. The following k equation is thus proposed for canonical shock/ turbulence interaction: is also modified in a similar way. 2a).4 is its high-Mach-number limiting value. There is no unique grid-independent solution of the standard k-ϵ model equations for the current shock–turbulence interaction. 12 in [18]). where the model proposed by Sinha et al. not included in Fig. Suppressing eddy viscosity. It is extensively used in engineering predictions of shock–turbulent-boundary-layer interaction. Sinha et al. 1). for example by a realizable model [23]. 1). The peak TKE value predicted by the different turbulence models. [16] reproduces LIA results closely. The overamplification of turbulent kinetic energy by the standard k-ϵ model is due to excessive production of turbulence at the shock wave. which are used here for model evaluation. 1 Evolution of turbulent kinetic energy k in the interaction of homogeneous turbulence with a normal shock at Mach 1. in the form of dilatational dissipation and pressure dilatation. they deteriorate model predictions in the upstream boundary layer and therefore are not preferred in SBLI flows [22]. They identify a damping effect of the unsteady shock oscillations on TKE amplification and develop a model for this effect. The model.

the amplification in ϵ increases monotonically with shock strength. respectively. and the double prime corresponds to Favre fluctuations. The model equations are then written in an alternate form to eliminate the nonconservative source terms. however. 2. 2013 | http://arc. future direction toward extending the conservative k-ϵ formulation to the simulation of complex high-speed flows is discussed. . the peak k and ϵ values obtained on fine grids are found to differ significantly from the analytical solution. (2) represents production of turbulence due to mean-flow gradients and is responsible for the amplification in k across a shock. contrary to the observations made. Standard k-ϵ Model and Its Variations where c1 ˆ 1. for example. The production term in the ϵ equation [Eq. Also. the shock-unsteadiness model. The last term corresponds to viscous and turbulent diffusion.2 ‡ 0. A. the source terms contain nonconservative derivatives of the flow variables. They are balanced by the source terms on the right-hand side. The jump in TKE therefore increases on successive grid refinement and does not reach a grid-converged value. and the nonconservative nature of the turbulent source terms are discussed next. and an additional parameter b1 of TKE due to unsteady shock motion.J052289 The realizable k-ϵ model. [16] propose a shock-unsteadiness correction to the k-ϵ model. Increase in the mean flow temperature and viscosity at the shock further adds to the ϵ-jump. as described previously.09. The effect of shock strength. The nonphysical behavior of the k-ϵ solutions are most prominent for strong shock waves and are probably caused by the nonconservative nature of the source terms in the governing equations. In some cases. (4). Neglecting dissipation in Eqs. Sinha et al. and the nonphysical behavior of the production terms is highlighted. The modified ϵ equation is given by u ~ ρ  ϵ2 ~ ρ ∂ϵ 2 ∂u  ϵ − c2 ˆ − c1 ρ ∂x 3 ∂x k (6) II. The physics-based shockunsteadiness k-ϵ model. grid sensitivity. in some cases. and variability due to changes in inlet conditions are investigated. The advantages of the new formulation over the traditional nonconservative k-ϵ equations are shown for the chosen test cases. g 0 02 is given by the Boussinesq The normal Reynolds stress u approximation: The closed-form solutions are plotted as a function of upstream Mach number and compared with LIA results in Fig. In particular. The evolution of k and ϵ across the normal shock wave is presented for a range of upstream mean flow Mach numbers. The only exception is for cases where ϵ is large. (5) and (6) and integrating the resulting equations analytically yields the following amplification in k and ϵ in terms of the mean density ratio across the shock wave:   2c  30 ρ k2 ˆ 2 1 ρ k1 and   2c  31 ρ ϵ2 ˆ 2 1 ρ ϵ1 (8)  ϵ2 ~ ρ ∂ϵ ϵ ∂u ∂  u 0 02 u ~ ˆ −c1 ρ − c2 ‡ ρ ∂x k ∂x ∂x k (3) where the overbar and tilde represent Reynolds and Favre averaging. and the second term corresponds to dissipation effect. compresswhere μT ˆ cμ ρ ibility corrections are not included in the k-ϵ equations.2514/1. The convection terms on the left-hand side of Eqs. and it has been discussed in previous work [18]. the turbulence transport equations in the standard k-ϵ model are presented. In this paper. As shown next. The closed-form solutions thus obtained are plotted in Fig. The upstream turbulence quantities correspond to the conditions for which DNS data are reported by Larsson and Lele [10]. The production ~ ∕∂x†2 and therefore takes very large values of k is proportional to …∂u at a shock wave. ~ 2 g 0 02 ˆ 4 μ ∂u u k −ρ − ρ 3 T ∂x 3 (4)  k2 ∕ϵ. Integration of the source term across the mean shock wave yields a contribution proportional to 1∕δ toward TKE amplification.AIAA Early Edition / SINHA AND BALASRIDHAR 3 Downloaded by UNIV OF TEXAS AUSTIN on June 10. which scale as ϵ Δk ∼ − δ and ~ u Δϵ ∼ −c2 ϵ2 δ ~ uk (7) The standard k-ϵ model [25] applied to the steady one-dimensional interaction of homogeneous isotropic turbulence with a normal shock wave takes the following form: ~ ∂k ∂ g 0 02 ∂u u ~ ˆ −ρ u ϵ ‡ ρ −ρ ∂x ∂x ∂x    μ ∂k μ ‡ T σ k ∂x    μT ∂ϵ μ ‡ σ ϵ ∂x (2) The mean shock thickness δ in a RANS solution is proportional to the grid size. The resulting k equation can be written as u ~ ρ ~ ∂k 2 ∂u ϵ k − ρ ˆ − c0 ρ ∂x 3 ∂x (5) where the effects of production and shock-unsteadiness damping are 0 combined into c0 ˆ 1 − b1 . and cμ ˆ 0. including the recent stress-limited k-ω model of Wilcox [24]. Here. and the corresponding discretization error attain large values in a flow discontinuity. strong shock waves. and they are discussed separately. 2 and can be assumed to hold in the inviscid limit or in highReynolds-number | DOI: 10. See [18] for additional details. and the k-ϵ model with μT ˆ 0 can be analytically integrated across the shock if the effect of turbulent dissipation is neglected in the region of the shock wave. It tends to saturate to a value close to 2 for high-Mach-number interactions.aiaa. The k amplification is close to unity for weak shock waves (M → 1) and increases with shock strength. The k-ϵ solution. They have negligible contribution at a shock wave and are therefore neglected. as observed for smooth solutions. On the other hand. δ is the mean shock thickness computed in a CFD simulation.21M and c2 ˆ 1. the error does not decrease in magnitude with successive grid refinement. This is. is subsequently described. the CFD solutions of these k-ϵ models should approach their respective exact integration limits as the grid is progressively refined. The second term in each equation represents turbulent dissipation and determines the decay rate of turbulence on either side of the shock wave. As noted earlier. we systematically study the numerical characteristics of the k-ϵ solution for canonical shock/turbulence interaction. Results are also presented for varying upstream values of the turbulence variables. and it decreases as the grid-point density at the shock is increased. the error can amplify on fine grids to yield unrealistic values of k and ϵ at the shock. The dissipation effect therefore decreases as the shock gets thinner on successive grid refinement. The eddy viscosity is set to zero in 0 is used to model the damping Eq. (3)] also exhibits a similar nonphysical trend.2 are model parameters. The first term on the right-hand side of Eq. Finally. The extension of the new conservative formulation to other turbulence models. An alternate conservative form of the k-ϵ equations is derived and implemented in the finite-volume code. do not show a converging trend with successive grid refinement. Governing Equations In this section. (5) and (6) represent the evolution of k and ϵ in the shock-normal direction. Effect of grid refinement on k and ϵ amplification at the shock is quantified. This is due to an amplification in enstrophy that contributes to the solenoidal dissipation rate. A finite-volume-based CFD code is used for the simulations. The first term results in turbulence amplification at the shock. Further. The turbulent dissipation terms cause reduction in k and ϵ at the shock wave.

A symmetric second-order discretization of the corresponding velocity derivative is used in the current CFD implementation. There is negligible variation in the ~ 2 ∂k 2 ∂u 2 ˆ − c0 ρk ρ−3c0 − ρϵρ−3c0 ∂x 3 ∂x (9) Mass conservation across the shock wave is used to write ~ ρuk ~ 2 ∂ −2c0 2 ∂u ρ 3 ˆ c0 ρk ρ−3c0 ∂x 3 ∂x (10) Adding Eqs. Other two-equation turbulence models like SST [26]. the net error is a function of the higher-order derivatives [Eq. The mean flow Mach number for the test cases range from low supersonic to nearly the hypersonic limit. Contrary to the of ∂3 u convective fluxes discussed previously. (5) and (6) can be written in a conservative form using the Reynolds-averaged mass conservation equation. The production terms in the standard k-ω equations have a form similar to those in Eqs. The flow variables are normalized by the mean density and the mean speed of sound upstream of the shock wave. (5) and (6) are presented next that eliminate the nonconservative source terms. Implicit time integration is achieved using the dataparallel line-relaxation method of Wright et al. A transport 2 equation for f can be derived by multiplying Eq. (5) by ρ−3c0 : ~ ρ−3c0 ρu 2 The Reynolds-averaged Navier–Stokes equations are solved for the mean flow. where the velocity gradients in the production terms are discretized using a central difference scheme akin to that presented in the Appendix. (9) and (10). example ρ we get a leading-order error in terms of the derivatives evaluated at the cell boundaries. (A8). contributions from adjacent cells cancel at the common interfaces. for u ~ k in Eq. Further. The positive and negative eigenvalues are used to split the Jacobian matrix into A‡ and A− . It is of the form h in Eq. and the code has been used successfully in several supersonic and hypersonic applications. where the turbulence model equations are fully coupled with the mean flow conservation equations [29]. outside the shock wave. This often results in nonphysical trends in k and ϵ amplification with successive grid refinement. Higherorder reconstruction is achieved by using the MUSCL approach. the overbar on ρ is dropped to avoid potential conflict with the sign of the exponent. The inviscid fluxes are computed with a modified. and the convective term on the left-hand side is conservative. B. The development of the k-ϵ equations presented previously can be easily extended to the k-ω model. It can therefore be transformed to an equivalent conservative form Eq. (5) and (6). These derivative terms take large values in the region of a flow discontinuity. and T −1 and T are the left and right eigenvector matrices. III. the previous equations can be cast into the following form: ~ ρu ~ ρu ∂f 2 ˆ −ρgρ3…c1 −c0 † ∂x (13) (14) ∂g ρg2 2…c1 −c0 † ˆ −c2 ρ3 ∂x f which are similar to the original equations Eqs. (5). and Shih et al. It is shown that the numerical results are different. these error terms from adjacent cells add up. Alternate forms of Eqs. The simulation results presented in the next section highlight several aspects of the nonconservative k-ϵ equations. The turbulent source terms are evaluated at the cell centers. Similar development for the ϵ equation leads to a new variable: g ˆ ϵρ−3c1 2 and the corresponding transport equation is . Thivet et al. The dissipation terms retain their 2 original form with an additional factor of ρ3…c1 −c0 † . The canonical shock–turbulence interaction flows computed in this work are listed in Table 1. However. (2) and (3).J052289 The convection terms in Eqs. and the two-equation k-ϵ model is used for turbulence closure [25].2514/1. Wilcox [24] presents a stress-limiter k-ω model. We thus get a well-behaved numerical solution at a flow discontinuity.725. Here. A finite-volume approach is used. For a normal shock. the convective fluxes are evaluated at cell interfaces. Some representative results are presented in [19. as well as the recent shock–turbulence k-ϵ model [18] can also be developed in a similar way. (A11). low-dissipation form of the Steger– Warming flux-vector splitting method [30]. (13) following the steps delineated previously. as discussed previously.20. The most energetic wave number κ 0 in the incoming turbulence field is taken as the characteristic length scale. (5) and (6) are nonconservative. 2013 | http://arc. The corresponding truncation error involves higher-order derivatives of the conserved quantity. see Eq. When all the cells spanning a shock wave are taken together. which limits TKE production at a shock. The flux vector is thus evaluated as a combination of the positive and negative flux contributions from the left and right side of a cell face i ‡ 1∕2: Fi‡1∕2 ˆ A‡ Ul ‡ A− Ur For a first-order flux evaluation. The exact integrated effect across the shock is thus recovered. [31]. Note that the quantity f is continuous across the shock. These derivatives scale as 1∕δ3 and result in large error in a shock wave. They also exhibit nonphysical turbulence amplification at a shock wave. (5) with c0 ˆ 1. the corresponding k equation can be written in the form of Eq. On the other | DOI: 10. we define 2 f ˆ kρ−3c0 2 that is constant across a shock. On integration over the finite-volume cell. The Jacobian A ˆ ∂F∕∂U is diagonalized as A ˆ T −1 ΛT where Λ is a diagonal matrix consisting of the eigenvalues of the system.4 AIAA Early Edition / SINHA AND BALASRIDHAR Downloaded by UNIV OF TEXAS AUSTIN on June 10. the production terms in Eqs. (A1) and the corresponding discretization error exhibits a telescoping effect at the cell interfaces. (8). we get ~ ρu ∂f 2 ˆ −ρϵρ−3c0 ∂x (11) where the source term with nonconservative derivative is eliminated. The corresponding error term involves integration ~ ∕∂x3 over the finite-volume cell Eq. Conservative Formulation ~ ρu 2 ∂g ρϵ2 −2c1 ˆ −c2 ρ 3 ∂x k 2 (12) Substituting ϵ ˆ gρ3c1 and k ˆ fρ3c0 . thinner shocks obtained on finer grids can lead to larger error than in coarse grid simulations. The viscous fluxes due to molecular and turbulent diffusivities are computed using a second-order central difference method applied at each cell face. from the exact integration Eq. except for the absence of the production terms. see Eq. The incoming turbulence field is assumed to be primarily composed of vortical fluctuations. (A10)] evaluated in a region of smooth flow solution. The flux vector F is evaluated at each cell face using the conserved variables vector U stored at cell centers. [28]. realizable k-ϵ models of Durbin [27]. [23]. In a finite-volume approach. (A5). as the leading error term vanishes on successive grid refinement. Their overall contribution across a shock wave therefore involves high-order velocity derivatives evaluated in the region of the flow discontinuity. Ul ˆ Ui and Ur ˆ Ui‡1 . Simulation Methodology Noting that k ∝ ρ3c0 across the shock wave in the inviscid limit.aiaa.32]. both qualitatively and quantitatively.

followed by a rapid decay behind the shock.6 times the upstream value) compared to the conservative form and DNS data. Further. the diffusive fluxes are not included in the computation because their contribution is expected to be small in the current shock–turbulence interactions. Both k and ϵ decay rapidly in the postshock flow to reach negligible values for x > 10. and turbulence level in the upstream flow. 1) The initial and boundary conditions for k and ϵ are transformed to the new variables f and g.6) stations. The conservation form of the model equations is incorporated by interpreting the turbulence variables in the code as f and g. 2013 | http://arc.15 × 10−4 . By comparison. For the strongest interaction (M ˆ 3. there is overamplification by an order in magnitude for the strongest . These include the shock strength in terms of the Figure 3 presents the evolution of TKE and turbulent dissipation rate for three test cases with varying shock strengths. grid resolution at the shock wave.22.57 × 10−2 2. slightly overpredicts the peak turbulent dissipation rate and is closer to the DNS data further downstream. 4a and 4b. This may be because of anisotropy effects that are prominent in the near field. These are extrapolated to the inlet station using the decay relations for homogeneous isotropic turbulence. (8)] because of the effect of turbulent dissipation in the shock region.5 2. predicted well by the conservative equations. In the near field (0 < x < 10).88 × 10−2 2.5. instead of k and ϵ.5 case.76 × 10−2 2.3 × 10−4 10. The nonconservative formulation. listed in Table 1.56 × 10−2 ϵ0 11. They match the DNS data downstream of the shock well.6) and exit (x ˆ 29. The same is true for the ϵ amplification presented in Fig. The solutions computed using nonconservative and conservative formulations of the k − ϵ turbulence model are compared with DNS data of Larsson and Lele [10]. ϵ is overpredicted. 4b. The prediction of turbulent dissipation rate matches DNS data at the shock and far downstream.5 case. are specified at the upstream boundary. and the turbulence variables are set to their inlet values in the entire domain. 3c and 3d.7 % compared to the conservative results) at the shock. 3) The original turbulence variables k and ϵ are recovered from the converged flowfield solution by invoking the reverse transform. the conservative formulation results for k and ϵ show the same qualitative trend as in the Mach 2. Availability of a closed-form solution for the current problem makes it ideal for testing different numerical methods and formulations. 2) Computation of the turbulent source terms are modified by dropping the production term and multiplying the dissipation term by 2 a factor ρ3…c1 −c0 †. Numerical Results and Comparison The canonical shock-homogeneous turbulence interaction test cases listed in Table 1 are computed using the traditional nonconservative form of the k-ϵ equations and the new conservative formulation presented previously. between the inlet (x ˆ −7.7 × 10−4 10. DNS solves the full Navier–Stokes equations including the viscous effects. The amplification in k and ϵ across the shock wave are also compared with the exact solution [Eq.5 2. where the turbulent Mach number immediately upstream of the shock wave is 0.69 × 10−2 2. given by Eq. The downstream decay is also reproduced well.AIAA Early Edition / SINHA AND BALASRIDHAR 5 Table 1 Mean and turbulent flow quantities for the interaction of homogeneous turbulence with a normal shock. except for an overprediction of the turbulent dissipation rate in the range 0 < x < 6.5 case.J052289 thermodynamics variables upstream of the shock wave.5) presented in Figs. As earlier.2514/1. Note that computations of the convection fluxes are not altered because they have the same form for the original and new turbulence variables. The numerical predictions are compared with the exact inviscid solution [Eq. The computational domain. see [18] for further details.3 × 10−4 10.75 3.64 × 10−2 2. The effect of grid refinement is discussed in detail in the following section. on the other hand.aiaa.9 × 10−4 9. and it results in a lower TKE downstream of the shock. the normalized values of k and ϵ correspond to the inflow station M 1. The turbulent dissipation rate is amplified by a factor of 47. and extrapolation condition is employed for the turbulence variables at the exit boundary. Minimal changes are then required to the CFD code to implement the conservative form of the k-ϵ equations [Eqs. In this case. The dissipation effect [Eq.5 4. and the discrepancy is larger than the Mach 2. denoted by Δ.84 yields k1 ˆ 0. Rankine– Hugoniot jump relations are used to obtain the back pressure. the conservative form of the model equations match the DNS amplification of k and ϵ in Figs. The conservative formulation also matches the ϵ amplification at the shock but underpredicts the DNS data in the postshock flow. The mean flow is initialized to a hyperbolic tangent profile. both the conservative and nonconservative models give comparable levels of the postshock turbulent kinetic energy.6. The amplification in k at the shock is. (2) and (3)]. (7). Both k and ϵ predictions are lower than the inviscid limit [Eq. This is opposite to the trend observed at lower Mach numbers and is caused by the high values of ϵ at the shock. On the other hand. (7)] is neglected while integrating the respective production terms to arrive at Eq. The amplification in k and ϵ obtained from DNS is therefore lower than the inviscid limit.7 k0 2.0 2. where the important physical and numerical parameters of the problem are varied independently. The peak ϵ level is grossly overpredicted (6.25 4. The normalized values of TKE and turbulent dissipation rate are thus calculated as k1 ˆ M2 t 2 and 5 M3 1 ϵ1 ˆ p t 3 Reλ κ 0 λ (15) A nondimensional value of the Taylor microscale κ 0 λ ˆ 0. the net turbulent dissipation of k in the region of the shock wave.0242 and ϵ1 ˆ | DOI: 10.8 × 10−4 DNS source Larsson and Lele [10] — Larsson and Lele [10] — — Larsson and Lele [10] — Larsson and Lele [10] upstream mean flow Mach number.66 × 10−2 2. the nonconservative formulation yields a higher amplification of TKE (by 12.0 3. (8)] obtained by direct integration of the k-ϵ equations in the inviscid limit. For the Mach 2. DNS data for a limited number of cases are presented by Larsson and Lele [10]. both k and ϵ values are normalized by their values immediately upstream of the shock wave.2 × 10−4 10. IV. Simulation results are documented next. The jump in TKE at the shock obtained using the nonconservative equations is found to be lower than that predicted by the conservative equations and DNS. The nonconservative formulation of the k-ϵ model predicts a very high ϵ in the Mach 3. Figure 4 plots the jump in k and ϵ across the shock as a function of the upstream mean flow Mach number. 3e and 3f. The error increases dramatically with Mach number.5 interaction. Effect of Upstream Mach Number Downloaded by UNIV OF TEXAS AUSTIN on June 10. the overprediction of the turbulence dissipation rate is due to the physical limitation of the linear-analysis-based k-ϵ model.61 × 10−2 2. Results are evaluated against available DNS data. Freestream and inlet conditions. The original CFD code is based on the conventional form of the turbulence model equations [Eqs. (8). Higher postshock ϵ results in a steeper decay of both TKE and turbulent dissipation rate in the downstream flow. and it is found to be close to the conservative RANS predictions in Figs. is large enough to cause a significant drop in the peak TKE level obtained in the interaction. The k amplification computed using the conservative procedure increases with Mach number and follows the same qualitative trend as the exact integration. A.4 × 10−4 10. (8)] and DNS data [10]. The Reynolds number based on Taylor microscale is 40 at this location. however. As earlier. is discretized using 200 equispaced points in the shock-normal direction. The nonconservative formulation of the k-ϵ turbulence model predicts a higher amplification of turbulent dissipation rate at the shock. These are listed next. (13) and (14). For an upstream Mach number of 1.0 × 10−4 9.

The TKE and dissipation rate computed using the conservative and nonconservative RANS codes are presented in Fig. 3 f) Variation of TKE and turbulent dissipation rate in shock/homogeneous turbulence interaction at a.aiaa.7) simulated. c. and a majority of the variations due to grid refinement are localized in the vicinity of the shock wave.5.2514/1. and 0. d) Mach 2. there is a complete mismatch between the qualitative variation of k2 ∕k1 with Mach number obtained using the traditional nonconservative form of the k-ϵ model and that expected from the direct integration of the equations. 5b.5. 5a. the postshock k level is 30% lower than the exact value. 0. For the strongest interaction (M ˆ 4.186.J052289 2 Nonconservative Conservative 4 Nonconservative 3 Conservative DNS k 1 DNS ε 2 1 0 -5 0 5 10 15 20 0 -5 0 5 10 15 20 x x c) d) 10 2 8 Nonconservative Conservative 6 Conservative 4 k 1 DNS ε DNS Non conservative 0 -5 2 0 5 x 10 15 20 0 -5 0 5 x 10 15 20 e) Fig.5. such that the shock-normal grid spacing for the three grids is 0. More | DOI: 10. on the other hand. B. f) Mach 3. The reason is explained earlier as a result of the competing effects of the production and dissipation terms in the k equation. There is negligible variation in the TKE amplification computed using the conservative formulation on the three successively refined . increases on successive refinement. the 1200-point grid yields essentially identical solution to that obtained using 400 points.5 test case is computed using three grids: the baseline grid of 200 points and two finer grids of 400 and 1200 points. the conservative code results show a converging trend.031. A similar trend is observed in Fig. The solutions follow the same qualitative trend as earlier. 5. Grid points are uniformly distributed between the inlet and exit locations. b) Mach 1. The peak value of ϵ in Fig. The TKE amplification. followed by a drop in the peak k value for strong shock waves. increases with Mach number for weak and moderately strong shocks. obtained using the nonconservative formulation. By comparison. 2013 | http://arc. The Mach 2. interaction (M ˆ 4.093. and e.7).6 2 AIAA Early Edition / SINHA AND BALASRIDHAR DNS 2 Nonconservative Nonconservative Conservative 1 k ε Conservative 1 DNS 0 -5 0 5 x 10 15 20 0 -5 0 5 x 10 15 20 a) b) 5 Downloaded by UNIV OF TEXAS AUSTIN on June 10. Grid Sensitivity The sensitivity of the results to the variation in grid-point density is studied next.

as given by Eq. In the conservative case. For strong shocks (M > 3). both k and ϵ amplifications approach the exact inviscid solution as the grid-point density is increased. Also. The deviation from the exact solution increases with shock strength. 6. (A9). This is because of a corresponding reduction in upstream turbulent dissipation rate ϵ1 in the simulation where k1 is decreased. (5) and (6) are of the form of Eq. A lower dissipation effect [Eq. where the amplifications of k and ϵ obtained using the three grids are plotted as a function of upstream Mach number. The ϵ amplification in | DOI: 10. These values are also listed in the table and are used to calculate the different Mach number cases presented previously. This is because the error term is proportional to the jump in mean velocity at the shock.2514/1. 6c obtained using the nonconservative formulation exhibits a diverging trend as the computational grid is refined. The TKE amplification obtained using the nonconservative formulation also increases with successive grid refinement. For the Mach 4. Overall.AIAA Early Edition / SINHA AND BALASRIDHAR 7 2. The value of TKE immediately upstream of the shock is increased by a factor of 2. the deviation of the numerical solution from the exact inviscid integration [Eq.7 case. The effect of turbulent dissipation at the shock scales as per Eq. The nonconservative solutions. 5 a) b) Evolution of TKE and turbulent dissipation rate in the Mach 2.5 Nonconservative 4 2 Conservative 0. Once again. The variation of TKE amplification with Mach number for higher and lower upstream turbulence level have the same qualitative variation as the baseline case. Next. and the turbulent dissipation rate increases by a factor of 8 to maintain the same value of Reλ as in the baseline simulation.5 200 points 400 points 1200 points 10 200 points 400 points 1200 points 2 8 Nonconservative 6 k Conservative 1 ε 1.5 test case for varying number of grid points in the shock-normal direction. The results obtained for different shock strengths using the conservative k-ϵ formulation are plotted in Fig. which results in a decrease in ϵ1 by a factor of 8.5 -2 0 2 4 x 6 8 10 0 -2 0 2 4 x 6 8 10 Fig. This can be explained by the 1∕δ3 ∼ 1∕Δx3 variation of the leading velocity derivative in the error term. the amplification in turbulent dissipation rate at a fixed Mach number increases with grid refinement. The grid sensitivity of the numerical solutions for varying shock strengths is presented in Fig. the location of the peak moves closer to x ˆ 0 as the shock gets thinner. and they decrease (as Δx2 ) as the grid spacing is reduced. The amplification is higher when the upstream TKE level is half of the baseline value. On the other hand. 2013 | http://arc. the trend is limited to weak and moderate shock strengths. (7) and is proportional to the shock thickness (δ ∼ Δx). C. (8)] vanishes as Δx → 0 on fine grids.J052289 b) Variation in k and ϵ amplification with upstream Mach number as computed using the conservative and nonconservative formulations. The turbulent p Mach number is thus enhanced p by 2. A 200-point grid is used in all the simulations. The corresponding values of k1 and ϵ1 are listed in Table 2. 4 Downloaded by UNIV OF TEXAS AUSTIN on June 10. 7. Effect of Upstream Turbulence Simulations are performed by varying the upstream turbulence levels. show an increase in TKE peak value as the grid is refined.5 Conservative DNS 5 Conservative 1 1 2 3 4 5 6 1 2 3 4 5 6 M M a) Fig. The discretization error in the convective terms in Eqs. Both the discretization error and the dissipation effect decrease with successive grid refinement. with the latter being the dominant contribution at small Δx.5 20 Nonconservative 2 Exact Nonconservative 15 k2/k1 ε2/ε1 Exact 10 DNS 1. resulting in a drop in the postshock TKE level with increasing Mach number. and the sensitivity of the results is analyzed. Further.aiaa. on the other hand. and it is lower for the higher upstream turbulence level. (7)] in the shock region results in a higher k2 ∕k1 that is closer to the exact inviscid result [Eq. compared to the baseline cases presented in Table 1. the finest grid yields a peak value that is more than 600 times the upstream level. (8)]. the high ϵ amplification across the shock dominates over the production effect. (A10) in the shock wave. the upstream TKE is reduced p by a factor of 2. meshes. a higher ϵ1 in the simulation with higher . 2.

The jump k2 ∕k1 in Fig. and its magnitude decreases with successive grid refinement. which prescribe constant values for the model parameters.22 0. the new formulation can be combined with the standard form of the turbulence model.5 1 2 3 4 5 6 1 1 2 3 4 5 6 M M a) b) 12 Downloaded by UNIV OF TEXAS AUSTIN on June 10. such that the conservation form is solved in high-compression regions and the original form of the equations is retained elsewhere in the flow domain. as seen earlier. 7b exhibits a trend identical to the TKE data. Also. 2). To apply the proposed conservative form of the turbulence model equations to general SBLI configurations. The deviation from the theoretical curve is due to the effect of turbulent dissipation in a finite-thickness shock wave.J052289 10 3 Exact 200 points 400 points 1200 points 10 Exact 200 points 400 points 1200 points 10 2 8 ε2/ε1 ε2/ε1 10 1 6 4 2 10 0 1 2 3 M 4 5 6 1 2 3 M 4 5 6 c) d) Fig. 8a attains highest value at M ˆ 3 for the simulation with upstream turbulence level half of the baseline condition. 8b. using a shock-identifier function . 6b and 6d) equations.0121 Mt 0.31 0.0242 0.23 × 10−4 Twice Baseline Half Interaction of shock waves with turbulent boundary layers involves additional gradients in the mean flow quantities. 6 Turbulence amplification for varying grid-point density computed using the nonconservative (Figs. TKE amplification for all simulations drop for M > 3. The current shock-unsteadiness model is applied to predict flow separation in SBLI flows in [20. whereas it is lowest for the case when incoming k value is increased. we need to address these and related issues. D. This has been achieved successfully for applying the shock-unsteadiness correction only in the vicinity of shock waves. Some ideas for generalization are discussed next.32].59 × 10−3 9. The results computed for reduced inlet TKE is again closest to the exact inviscid solution. In the presence of gradient-based production. the conservative k-ϵ equations give accurate and physically consistent solution for the shock/homogeneous turbulence interactions considered in this work.15 × 10−4 3. Thus.8 2. 6c. Note that a log scale is used in Fig. The variation of shock strength in the interaction region is computed in terms of the local upstream Mach number normal to the shock wave. The foregoing results show that the conservative form of the k-ϵ turbulence model is able to predict shock–turbulence interaction consistently over a range of upstream Mach numbers. Once again. It is shown that model parameters based on an average shock strength can yield satisfactory results.16 ϵ1 2. Also. The nonconservative simulations also exhibit an increase in turbulence amplification across the shock as the upstream turbulence level is decreased. The same is true for other turbulence models (like the stress-limited k-ω model). which shows large changes in ϵ amplification at high Mach numbers. 2013 | http://arc.5 Exact 200 points 400 points 1200 points Exact 200 points 400 points 1200 points 2 2 k2/k1 1. The amplification in turbulent dissipation rate across the shock in Fig.aiaa. the shock is usually oblique to the incoming flow and may be curved due to Mach number variation in the boundary layer.5 AIAA Early Edition / SINHA AND BALASRIDHAR 2. k1 shows a larger deviation from the inviscid limit. The effect of varying upstream turbulence level is most prominent in Fig. 6a and 6c) and conservative (Figs. The turbulence model equations can thus be transformed into equivalent conservation form using representative average values of c0 and c1 . The RANS solution matches DNS data well. Generalization to Shock–Boundary-Layer Interaction Flows Table 2 Turbulence level immediately upstream of the shock wave for varying inlet conditions k1 0. and vice versa. and discrepancies in the postshock turbulence evolution in the near field is due to limitations in the physical | DOI: 10. the predicted turbulence amplification across the shock closely follows the analytical solution. which is almost identical to the linear theory predictions (see Fig. a higher jump ϵ2 ∕ϵ1 is observed for lower upstream ϵ1 . An appropriate blending can be devised.0484 0.2514/1.5 1 k2/k1 1.

5 Exact k1 2k1 k1/2 2 150 Exact k1 2k1 k1/2 100 k2/k1 1. The numerical predictions are compared with available direct numerical simulation (DNS) data and results of linear interaction analysis. ∂h hi‡1∕2 − hi−1∕2 ≃ Δx ∂x (A3) An upwind-biased method is usually employed to compute the fluxes hi‡1∕2 and hi−1∕2 . which eliminates the nonconservative source terms. A1). Here. It is shown that the nonconservative source terms in the k-ϵ equations can result in nonphysical solution for strong shock waves. Conclusions The paper presents application of the k-ϵ turbulence model to the interaction of homogeneous isotropic turbulence with normal shock waves. An alternate formulation of the k-ϵ equations is proposed. Using Taylor series to expand the right-hand side gives  ∂h ∂x  ‡ i Appendix A This appendix presents the discretization error for an equation of the form   1 Δx2 ∂3 h ‡ ::: 3 8 ∂x3 i (A4) where the second term is the leading-order truncation error. 2013 | http://arc. for example ρ term. which can cause rapid decay of the downstream turbulence. There is excessive amplification of turbulent dissipation rate at high Mach numbers.5 ε2/ε1 50 1 1 2 3 4 5 6 1 2 3 4 5 6 M M a) b) Fig.aiaa. in the region of a flow discontinuity.AIAA Early Edition / SINHA AND BALASRIDHAR 9 2. (5). we get . the prediction of the conservative k-ϵ equations approaches the exact inviscid solution and linear theory results with successive grid refinement. The convective term on the left-hand side is discretized in terms of the fluxes evaluated at the cell | DOI: 10.5 Exact k1 2k1 k1/2 2 10 8 Exact k1 2k1 k1/2 k2/k1 ε2/ε1 1. h represents the convective u ~ k in Eq. [19].J052289 b) Effect of varying inlet turbulence level on k and ϵ amplification computed using the conservative k-ϵ equations. 2. On integration over the finite-volume cell. The new formulation thus yields significant improvement over the conventional turbulence model equations and forms the basis for further application to shock– boundary-layer interaction flows. Also. 8 Amplification in TKE and turbulent dissipation rate across a shock wave as a function of upstream Mach number for varying inlet turbulence levels computed using the nonconservative k-ϵ equations. like the production in the k equation: ~ 2 ∂u k ψ ˆ − c0 ρ 3 ∂x (A2) The convective and production terms in the ϵ equation can also be cast in this form.5 1 1 2 3 4 5 6 6 4 2 1 2 3 4 5 6 M M a) Fig.2514/1. A similar approach can be used here for the conservative formulation. For the ith cell (see Fig. 7 Downloaded by UNIV OF TEXAS AUSTIN on June 10. The corresponding results match DNS amplification of turbulent kinetic energy and its dissipation rate. ∂h ˆψ ∂x (A1) V. and ψ is a nonconservative source flux.

1993. 349–370. the first term represents the integrated production of TKE in the cell. and Zeman. the discretized source term [Eq. 160.. 1998.3275856 [11] Durbin.. F.. 334. pp. [2] Horstman. 39.1017/S0022112097004576 [9] Jamme.. K. 2003. Yan. 1. “Prediction of Hypersonic Shock-Wave/Turbulent Boundary-Layer Flows... B. J. 2009. 251.002 [7] Lee. 121–184. (A5)] for all the cells between xup and xdn gives . pp.J052289 . and Chassaing. P. AIAA Paper 1993-0200. A. 2002. 3. and the magnitude of error decreases monotonically as the grid is refined. [4] Barakos. x where the nonconservative error term involves higher-order derivatives evaluated in the region of the shock wave... “The Influence of Entropy Fluctuations on the Interaction of Turbulence with a Shock Wave. (A8)] for all the cells add up to 2 − c0 3 Z xdn xup k ρ ~ ∂u 2 Δx2 dx − c0 3 ∂x 3 Z xdn xup k ρ ~ ∂3 u dx − : : : ∂x3 (A11) i-1/2 up .. March 1997. Vol. “Direct Numerical Simulation of Canonical Shock/Turbulence Interaction.. A. Moin. The source term ψ involves nonconservative derivative of the shock-normal flow velocity.. P. Nos. G. .paerosci. P. S...1023/A:1021197225166 [10] Larsson. doi:10.1017/S0022112093003258 [13] Mahesh. Torres. Vol. . 353–379. 641–665. Jan. K. 126101. 21. P.1016/j. Vol. K. K.2006. Nos. In some cases (for example. S.” Progress in Aerospace Sciences. Dec. pp. N. Vol.aiaa. N.. Vol. T. Acknowledgments    2   Δx2 ∂2 h ∂ h ‰hi‡1∕2 − hi−1∕2 Š ‡ − ‡ : : : (A5) 24 ∂x2 i‡1∕2 ∂x2 i−1∕2 where the integrated error is a function of the derivatives evaluated at the cell interfaces. Vol.1017/S0022112093003519 [8] Mahesh. 2–3.. A1 Finite-volume grid for discretization of fluxes and source terms in the region of a flow discontinuity.. C. P. Vol. K.. and the second term is the discretization error. Integration over the finite-volume cell gives 2 − c0 3 Z xi‡1∕2 xi−1∕2 k ρ ~ ∂u 2 Δx2 dx − c0 ∂x 3 3 Z xi‡1∕2 xi−1∕2 k ρ ~ ∂3 u dx − : : : ∂x3 (A8) Here.1017/S0022112092002404 [12] Cambon.. and Drikakis. as per Eq. Similarly.” AIAA Paper 1992-0436. References [1] Coakley. i+1/2 Downloaded by UNIV OF TEXAS AUSTIN on June 10. dn i+1 i-1 i Fig. C.. and Lele. The derivatives attain large values. doi:10. [3] Huang.1016/S0376-0421(02)00069-6 [6] Roy. doi:10. and Moin. Dept. P..” Journal of Fluid Mechanics.” Journal of Fluid Mechanics. 155–174. 227–268.. doi:10. and the integrated error remains finite as the grid is refined. Coleman. “The Interaction of a Shock Wave with a Turbulent Shear Flow. “Rapid Distortion Analysis and Direct Simulation of Compressible Homogeneous Turbulence at Finite Mach Number. “Direct Numerical Simulation of Isotropic Turbulence Interacting with a Weak Shock Wave. pp. “Assessment of Various Low-Reynolds Number Turbulence Models in Shock-Boundary Layer Interaction. of (A7) where Taylor series expansion is used again to identify the leadingorder error. A1).. Sept. and Moin. such that the derivative ~ Δu ~ Δu ~ ∂u ∼ ∼ ∂x Δx δ where Δu ~ is the jump in the flow variable across the shock. doi:10.1063/1. 533–562. and Zheltovodov. “Advances in CFD Prediction of Shock Wave Turbulent Boundary Layer Interactions. Vol. AIAA Paper 1987-1367. pp. 68. H. strong shock waves). 1993. 242. The higherorder derivatives scale as ~ Δu ~ ∂2 u ∼ ∂x2 δ2 and ~ Δu ~ ∂3 u ∼ ∂x3 δ3 (A9) and are therefore unbounded in the limit Δx → 0.10 AIAA Early Edition / SINHA AND BALASRIDHAR u Shock ‰hdn − hup Š ‡ Δx2 24  ∂2 h ∂x2  dn −  ∂2 h ∂x2   ‡ ::: up (A10) where the leading error term involves derivatives evaluated outside the region of discontinuous solution and is therefore small in magnitude. N.” Journal of Fluid Mechanics.” Thermosciences Division. J. Δx is the grid spacing. O. Plasma Dynamics and Lasers Conference. K. S. 469–530. 1992. A symmetric second-order discretization of the velocity derivative yields   ~ −u ~ i−1 u 2  k i‡1 ψ ≃ − c0 ρ 3 2Δx      ~ ~ 2 ∂u Δx2 ∂3 u k ‡ ‡ : : : ˆ − c0 ρ 3 ∂x3 i 3 ∂x i (A6) The authors wish to acknowledge Johan Larsson of University of Maryland for providing the direct numerical simulation data for model evaluation. pp. 257. Cazalbou.” 31st Aerospace Sciences Meeting and Exhibit. S. Lele. “Direct Numerical Simulation of the Interaction Between a Shock Wave and Various Types of Isotropic Turbulence. This is not true in regions of flow discontinuity..” Physics of Fluids. 1993.12. D. Vol. doi:10. G.” AIAA 19th Fluid Dynamics. 42. No.1016/S0045-7825(97)00291-0 [5] Knight. P. G. F.. 7–8. G. C.. and Lele.2514/1. and δ is the computed shock thickness. Jan. and Blottner... June 1987. The total error accrued in the region of a shock wave can be estimated when the contributions of all the cells spanning the shock are taken together (see Fig.” Progress in Aerospace Sciences. “Rapid Distortion Theory for Homogeneous Compressed Turbulence with Application to Modelling. pp. The shock thickness computed using a shock-capturing method decreases with successive grid refinement. No. pp.. | DOI: 10.. T. (A9). doi:10. D.. . S. No. doi:10. J. it can be comparable or larger than the production term in magnitude.” Computer Methods in Applied Mechanics and Engineering.” Journal of Fluid Mechanics. “Review and Assessment of Turbulence Models for Hypersonic Flows.. G. 1992. . A. 2013 | http://arc. J. p. Lele.. and Coakley. then the derivatives are finite. G.. 2006. “Turbulence Modeling for High Speed Flows. doi:10. 12. J. Turbulence and Combustion.” Flow. S. Summation of the conservative flux terms [Eq. and Mansour. and Huang. If the solution is continuous. C. Panaras. “Turbulence Modeling for Complex Hypersonic Flows.

” Journal of Fluid Mechanics. TF-69. The Univ. 24.B34191 Sarkar. pp. “Analytical Linear Theory for the Interaction of a Planar Shock Wave with an Isotropic Turbulent Vorticity Field... 1–2.. 3. 135–150. 34. 2823–2838.. R. 2008. 2012. pp.” 23rd International Symposium on Shock Waves. [24] Wilcox. doi:10.. G. doi:10. 46.. Stanford Univ. A. Rept. C. “Simulation of Hypersonic Shock/Turbulent Boundary-Layer Interactions Using Shock-Unsteadiness Model.. and Zhu. pp. A. Vol. pp. 1995. “Importance of Limiting the Turbulent Stresses to Predict 3D ShockWave Boundary-Layer Interactions. 2290–2297. doi:10... and Candler. A. 9. doi:10. “The Analysis and Modelling of Dilatational Terms in Compressible Turbulence. No. 8. Vol. “Convergence Improvement of TwoEquation Turbulence Model Calculations. No. No.” International Journal of Computational Fluid Dynamics. p. V. 1989. No. P. doi:10. Stanford. doi:10..2514/1.1588306 Veera.. K. No.” AIAA Journal. A.1016/0142-727X(95)00073-Y [28] Shih. “Modeling the Effect of Shock-Unsteadiness in Shock/Turbulent Boundary-Layer Interactions.2514/3. W. 21. 227. K. 2005.. and Bonnet. R.. Wouchuk.. 1996. doi:10. 5. and Bose. C.” Computers & Fluids. Alem. Y. Turbulence Modeling for CFD.. Vol. No. “On the k-3 Stagnation Point Anomaly. M. D. 2012.8611 Pasha. doi:10.aiaa. 236–239...” AIAA Journal. K.. Z. doi:10.. pp. Mahesh. No. F. F. “Experimental Study of a Normal Shock/Homogeneous Turbulence Interaction.J052289 [19] [20] [21] [22] Mechanical Engineering. 8. Nos. 707. J.2514/2.3073744 Sinha. “Shock-Unsteadiness Model Applied to Oblique Shock-Wave/Turbulent Boundary-Layer Interaction. “Data-Parallel Line Relaxation Method for the Navier–Stokes Equations. D.066315 Barre.. G. C. and Candler. Vol. “A New k-ϵ Eddy Viscosity Model for High Reynolds Number Turbulent Flows..” Computers and Fluids. J. R. 46–60. 28. V. Erlebacher. A. J.586 [32] Pasha. V. 501–505.. 025101.1017/jfm. Vol. Vol.” Journal of Fluid | DOI: 10.36541 [25] Wilcox. Huete Ruiz de Lira.2514/3. 43.1016/0045-7930(94)00032-T [29] Sinha. A. G. 1. 15. Sept. “Computation of Shock Wave Turbulent Boundary Layer Interactions Using a Two Equation Model with Compressibility Corrections. 1998. P. Vol. and Friedrich. and Sinha. No. June 1998. “The Solution of the Navier– Stokes Equations Using Gauss–Siedel Line Relaxation.” International Journal of Heat and Fluid Flow... pp. D.. 2761. O.12149 [27] Durbin. La Canada. pp.. 968–974. Yang. 32. [26] Menter. 569–582. 17. pp. pp. 1. [30] MacCormack.1063/1. and Maksimov.265 Sinha. 2009. and Candler. 17. Mahesh. W.1007/BF01082582 [23] Thivet. doi:10. 36.” AIAA Journal. p.2012. V. pp.2514/1. No.. V. CA. M. H.” Applied Scientific Research. p. doi:10.. 3. 1603–1609.1080/10618560802290284 G. doi:10. A. Hussaini.1016/0045-7930(89)90012-1 [31] Wright. 066315. Vol. June 1991.2514/1.. pp. Shabbir. No.. Blaisdell Associate Editor .AIAA Early Edition / SINHA AND BALASRIDHAR 11 [14] [15] [16] [17] [18] Downloaded by UNIV OF TEXAS AUSTIN on June 10. K. “Modeling ShockUnsteadiness in Shock/Turbulence Interaction.79.” AIAA Journal.1063/1. and Velikovich.” Physics of Fluids.. K. 2003.1103/PhysRevE.... and Sinha. 11. Knight. S. pp. “Formulation of the k-ω Turbulence Model Revisited. V. W. 74–110. 1993... and Sinha. T. K. 2001. of Texas at Arlington. DCW Industries. A. 1. 227–238. A.. Vol. D. J.. F. Liou. Vol. 6. pp. 586–594. 89–90.. A. 1994.” Journal of Propulsion and Power. No. Candler. Vol. D. doi:10.” Physics of Fluids. 2009. D. K. doi:10. Vol.. 22. “The Evolution of Enstrophy in Shock/Homogeneous Turbulence Interaction. 2008. Vol. H..1017/S0022112091000204 Haidinger.13175 Sinha. 2013 | http://arc. 2. No. 1996.” AIAA Paper 1998-2649. “Modeling the Effect of Upstream Temperature Fluctuations on Shock/homogeneous Turbulence Interaction. No. G. doi:10. pp.. CA. and Candler. and Kreiss. 51. Vol. 8. G. K. 79.. K.. A..2514/1.. G. 473– 493. pp. K. S. 1598–1605. doi:10.” AIAA Journal. 1996.” Physical Review E. Vol. Zheltovodov.. 1998. “Two-Equation Eddy-Viscosity Turbulence Models for Engineering Applications. I. Inc. Vol.

Sign up to vote on this title
UsefulNot useful