Acta Astronautica 210 (2023) 372–383

Contents lists available at ScienceDirect

Acta Astronautica
journal homepage: www.elsevier.com/locate/actaastro

Compressible correction for separated and shear flow based on

structural compressibility
Chen Yi, Zhen Zhang, Jinping Li, Fanzhi Zeng, Chao Yan *
National Key Laboratory of Computational Fluid Dynamics, Beihang University, Beijing, 100191, China

A R T I C L E I N F O A B S T R A C T

Keywords: In this paper, different compressible correction strategies for k-ω turbulence model are proposed. The dilatation
Compressible flow dissipation term, which is significant in compressible flow, is taken into account and modeled in the Favre-
Computational fluid dynamics averaged k-equation and ω-equation. We propose correction model for the dilation dissipation term based on
Compressible corrections
the gradient Mach number in this paper, the coefficients of which are calibrated from experimental data on
Turbulence models
compressible turbulent mixed layers, and compare it to the Wilcox, Sarkar, and Heinz correction models. In
slope-cavity flow and oblique shock wave/boundary layer interactions, the performance of these corrections is
evaluated and compared to experimental data. The new correction has a similar or even better effect in sepa­
ration prediction, heat flux prediction, and other aspects, and the uncertainty of model form and coefficients can
be investigated further.

summarized the sources of uncertainty in numerical simulations, where

one of the main aspects of uncertainty in CFD systems is the uncertainty
1. Introduction of the model form, that is, the use of different assumptions to construct
turbulence models will have a large impact on the numerical simulation
In recent years, with the increasing attention and research on hy­ results. Turbulence models frequently used are derivated from the
personic vehicles design, it is a challenging problem for the designer to incompressible forms of Navier-Stokes equations, which limits their
accurately predict wall pressure and wall heat flux rate by means of CFD. ability to simulate compressibility of flows at high Mach number. When
There are three main categories of turbulence simulation methods: the Mach number is small (generally smaller than 5), the variation of
Direct Numerical Simulation (DNS)[1], Large Eddy Simulation (LES)[2] mean density and fluctuating density is relatively negligible with respect
and Reynolds-averaged Navier–Stokes (RANS)[3] method. Under re­ to the value of mean density, so that the compressible effect only works
strictions of the current development of computer hardware and con­ on the mean velocity and scalars[6]. In this case, there are rarely dif­
siderations of economic costs, The RANS method is still widely used for ferences in compressible and incompressible turbulent structures, which
the simulation of turbulence boundary layers. RANS methods are makes the Morkovin hypothesis appropriate and incompressible turbu­
divided into two major categories, namely, eddy viscosity models (EVM) lence models useable directly. However, as the Mach number increases,
and Reynolds stress models (RSM), where the eddy viscosity models are the compressible effect will have an impact on velocity and scalar
constructed based on the Boussinesq stress-strain hypothesis. Tradi­ fluctuations, as well as the structure of turbulence boundary layers. For
tional eddy viscosity models often show deficiencies, such as over­ example, for heat transfer calculations in combustion flows and for
estimating eddy viscosity at separated vortices and failing to reflect fluid estimating the spreading rate of the mixing layer in compressible shear
anisotropy at corner structures, due to the plausible Boussinesq hy­ flows, turbulence models constructed on the basis of the Morkovin hy­
pothesis[4]. The eddy viscosity model, on the other hand, is still the pothesis often lead to erroneous results. Under these conditions, the
mainstay of engineering calculations today due to its simple construc­ compressible effect cannot be overlooked, and origin turbulence models
tion, economy, and robustness. Therefore, the study of compressible are insufficient to account for the likelihood of mixing layer thickness
correction of turbulence model in this paper is based on the eddy vis­ overprediction.
cosity model . The most direct way to take into account the compressible effect is
In the present CFD advancement, quantifying CFD uncertainty has modifying the primary incompressible non-conservation equation into
become a major research priority and challenge. Schaefer et al.[5]

* Corresponding author.
E-mail addresses: [email protected] (C. Yi), [email protected] (C. Yan).

https://doi.org/10.1016/j.actaastro.2023.05.045
Received 13 January 2023; Received in revised form 24 April 2023; Accepted 30 May 2023
Available online 1 June 2023
0094-5765/© 2023 IAA. Published by Elsevier Ltd. All rights reserved.
C. Yi et al. Acta Astronautica 210 (2023) 372–383

Nomenclature ξ∗ , Mt0 , C1 , C2 , C3 = model parameters of compressible corrections

F1 = near-wall protection function
ρ, ui , p = density, velocity vector, pressure d = distance between the center of grid and the wall
h, E, q = specific enthalpy, total energy, heat flux β∗ , γ, β, σ k , σ ω = Model parameters of k-ω model
τlij , τtij = molecular viscosity, turbulent eddy viscosity k-ω Standard = Standard k-ω model
k = turbulence kinetic energy k-ω Wilcox = k-ω model with Wilcox’s correction
ω = specific dissipation rate k-ω Sarkar = k-ω model with Sarkar’s correction
ε = dissipation rate k-ω Heinz = k-ω model with Heinz’s correction
εs = solenoidal dissipation rate k-ω NCC = k-ω model with new compressible correction
εd = dilatation dissipation rate SWTBLI = shock wave/turbulent boundary-layer interactions
μ, μt = molecular viscosity, turbulent eddy viscosity RANS = Reynolds averaged Navier-Stokes
Sij , S = strain tensor and magnitude EVM = Eddy viscosity models
Ωij , Ω = vorticity tensor and magnitude RSM = Reynolds stress models
c = local speed of sound LES = Large Eddy Simulation
lt = turbulent length scale DNS = Direct Numerical Simulation
Mt , Mg = turbulent Mach number, gradient Mach number

compressible conservation equations. Catris et al.[7] fixed the diffusion ( )

terms in Spalart-Allmaras and different variants of two-equation models ∂ρ ∂ ρũj
+ =0 (2)
by involving the density variations in order to make them consistent ∂t ∂xj
with the logarithmic law in compressible boundary layer. However, the ( )
calculation indicated that the above correction is inadequate to predict ∂(ρ̃ui ) ∂ ρ̃ui ̃uj ∂p ∂ ( ̃l )
+ =− + τij − ρu′′i u′′j (3)
the compressibility effect in results. At present, the widespread approach ∂t ∂xj ∂xi ∂xj
is researching on Favre-averaged NS equations. There are four addi­
∂ [ ( ̃l ) ]
tional terms in Favre-averaged turbulent kinetic energy equation ̃
∂(ρE)
+
∂ [ ̃
(ρE + p)̃
]
uj = ui τij − ρu′′i u′′j − qj − ρu′′j h′′
̃ (4)
compared to the previous incompressible version, which are dilatation ∂t ∂xj ∂xj
dissipation, pressure dilatation, pressure diffusion and pressure work
[8]. Mansour et al. implied that[9] the pressure diffusion term is rela­ where uj , p, τij , E, q, h represent j-direction components of velocity,
tively small compared to the molecular viscous diffusion term and tur­ pressure, viscous stress tensor, total energy, heat flux and enthalpy.
bulent diffusion term by direct numerical simulation data. Grasso et al.
have researched[10] on hypersonic flat plate boundary layer flows, 2.2. Turbulence model
indicating that the pressure work term is negligible. Based on the above
two views, the most popular compressibility corrections currently are The base turbulence model taken in this paper is Wilcox k-ω two-
different closures of dilatation dissipation term and pressure dilatation equation model (1988)[8]. The original k-ω model can be written as
term. [ ]
In this paper, the application of a compressible correction model
∂ρk ∂ρui k ̃ ∂ ∂k
+ = Pk − ρβ∗ kω + (μ + σk μt ) (5)
based on structural compressibility to turbulence models is preliminarily ∂t ∂xi ∂xi ∂xi
considered. At the same time, due to the excellent performance on [ ]
∂ρω ∂ρui ω γ ̃ ∂ ∂ω
simulation of near-wall flows and aerodynamic heat, the Wilcox k-ω + = Pk − ρβω2 + (μ + σω μt ) (6)
∂t ∂xi υt ∂xi ∂xi
model (1988)[11], which is widely used in turbulent flow simulation, is
selected, and various compressible corrections invented so far are
where υt = μt /ρ, the eddy viscosity can be obtained by
considered. These corrections are evaluated for reattaching compress­
ible turbulent shear layer flow in slope-cavity and shock wave/turbulent k
(7)
μt = ρ
boundary-layer interactions (SWTBLIs) generated by impingement of an ω
incident oblique shock wave on a flat-plate boundary layer.
the turbulence kinetic energy production P
̃k can be written as

2. Turbulence models and compressibility corrections ̃k = min(Pk , 10ρβ∗ kω), Pk = − τt ∂ui

P ij (8)
∂xj
2.1. Governing equations
where
For flows in which the compressible effect should be taken into ac­ (
1 ∂uk
)
2
count, we must use the density-weighted averaging process suggested by τtij = ρui uj = − 2μt Sij − δij + ρkδij (9)
3 ∂xk 3
Favre[8]. Favre averaging is defined as φ = φ̃ + φ′′ , where φ
̃ and φ′′
represent the density-weighted averaged part and the fluctuating part, Sij is the strain rate tensor and δij is the Kronecker delta.
and ( )
1 ∂ui ∂uj
Sij = + (10)
̃ = ρφ/ρ
φ (1) 2 ∂xj ∂xi

where ρ represents the density, φ represents the scalars and “~” repre­ where the coefficient β∗ = 0.09, γ = 5/9, β = 0.075, σk = 0.5, σω = 0.5.
sents Favre-averaged variable, “-” represents time-averaged variable. The TKE production Pk can be expanded as
After Favre averaging process, we can derivated the Favre-averaged ( )2
Navier-Stokes equations of mass, momentum and energy as follows 2
Pk = μt S2 − μt
∂uk 2 ∂uk
− ρk (11)
3 ∂xk 3 ∂xk

373
C. Yi et al. Acta Astronautica 210 (2023) 372–383

where S is the scalar invariant of the strain rate Sij . ̃ t )εs

εd = ξ∗ F(M (18)
( )
∂ui ∂uj ∂ui
S2 = 2Sij Sij = + (12) ̃ t ) = (1 − F1 )F(Mt )
F(M (19)
∂xj ∂xi ∂xj
In most flows, S ≈ Ω, so the strain form of Pk can be simplified to the F1 is blending function and F(Mt ) is function of Mt , and different
vorticity form as approximations are as follows:

Pk = μt Ω2 (13) 1) Sarkar’s model[14]:

where Ω is the magnitude of the vorticity F(Mt ) = Mt2 , ξ∗ = 1 (20)

( ) ( )
∂ui ∂uj ∂ui 1 ∂ui ∂uj
Ω2 = 2Ωij Ωij = − , Ωij = − (14) 2) Wilcox’s model[8]:
∂xj ∂xi ∂xj 2 ∂xj ∂xi
( )
Ωij is the rotation tensor. It is an appropriate approximation of Pk F(Mt ) = Mt2 − Mt02 Н(Mt − Mt0 )
(21)
which can avoid some problem caused by numerical stability when ξ∗ = 2, Mt0 = 0.25
using the exact form of Pk in flows of high Mach number. The exact form
Considering that the compressible correction will adversely affect
and the vorticity form of the turbulence kinetic energy production term
the calculation of the near-wall boundary layer, the compressible
differ in form and assumptions, but despite this, the difference between
correction in this paper introduces the near-wall protection function F1
the two calculations is not significant in a larger number of flows, and
proposed by Brown[15]. The F1 function has been applied in the SST
none of the cases in this work are sensitive to the difference between the
model by [16]; and the function can well determine the near-wall region
two. In this work, the vorticity form is used empirically while taking the
in the turbulent boundary layer. The formula of protection function F1
criterion of numerical stability into consideration.
can be written as follow
{{ ]}4 }
2.3. Compressibility corrections [ ( √̅̅̅
k 500μ
)
4ρσω k
F1 = tanh min max ∗ , 2 , (22)
β ωd d ρω CDkω d2
For sake of compressible boundary layer flow, Favre averaging is
taken in instantaneous Navier-Stokes equation, while the density and
where d represented the distance between the center of grid and the
pressure are still time-averaged, we can get the transport equation of
wall,
turbulence kinetic energy as
( )
2ρσ ω ∂k ∂ω − 10
∂(ρk) ∂(ρ̃ui k) ∂̃ui ∂u′′ CDkω = max , 10 (23)
+ = − ρu′′i u′′j − τij i + ω ∂xi ∂xi
∂t ∂xi ∂xj ∂xj ∂p
(15)
∂
(
1 ′
)
′ ∂u ′′ ∂xi In hypersonic flow, the compressible effect including dilatation
u′′ τ − ρu′′j u′′i ui − p u′′j + p i − u′′i
∂xj j ij 2 ∂xj compressible effect and structural compressible effect. The dilatation
compressible effect is mainly caused by the non-zero divergence of ve­
On the right-hand side of equation (17), the first term is production locity in the flow field, the structural compressible effect caused by
caused by the correlation of Reynold stress and mean gradient of ve­ structural skewness of velocity in the flow field. In contrast to turbulent
locity. The second represents the dissipation of turbulence kinetic en­ Mach number Mt , the gradient Mach number Mg can reflect the struc­
ergy, which can be modeled as tural compressible effect and well distinguish the strong compressible
effect in mixing layer and the weak compressible effect in boundary
∂u′′i
τij = ρε = ρεs + ρεd (16) layer[17]. The gradient Mach number Mg , which is the ratio of the
∂xj
acoustic time scale to the flow distortion time scale, can be calculated by
√̅̅̅̅̅̅̅̅̅̅̅̅
where ε represents the dissipation rate, εs represents the solenoidal Mg = Slt /c (S = 2Sij Sij , Sij represent the deviatoric strain rate tensor,
′ ′ ′

dissipation which is same as the dissipation rate in incompressible TKE √̅̅̅

lt = k/ω represent the turbulent length scale and c represent the local
equation and εd represents the dilatation dissipation rate which is speed of sound)[18]. DNS results of supersonic mixing layer flow show
related to the compressible effect. The third term is diffusion term, in that structural compressibility has a greater impact on the turbulence
which from left to right are viscous diffusion, turbulent diffusion and model than dilatation compressibility. Through DNS data fitting, [19]
pressure diffusion respectively. The fourth term is pressure dilatation obtained that the model constant Cμ in the two-equation model is no
and the last term is pressure works. Contrary to equation (5), there are
longer a constant, but satisfies the following formula:
four additional terms in Favre-averaged TKE equation, which are dila­ ( )
tation dissipation, pressure diffusion, pressure dilatation and pressure Cμ = 0.07 exp − 0.4Mg (24)
work. In general consideration, the pressure diffusion is often modeled
together with the other diffusion terms, and the pressure work is ignored When the flow tends to be incompressible, i.e., Mg = 0, Cμ = 0.07 which
since it is neglected. Compressibility corrections we use frequently are may go against Cμ = 0.09 in standard k-ω, k-ε model as well as SST
models of dilatation dissipation and pressure dilatation[12]. model. Therefore, the Heinz compressible correction formula adopted in
A number of researchers has established modeling methods based on this paper is:
turbulent Mach number Mt , which can be calculated by ( )
Cμ = 0.09 exp − 0.4Mg (25)
√̅̅̅̅̅
2k (17)
Mt = This work presents a broad form of compressible correction based on
c
the gradient Mach number Mg , based on the characteristics that the
where k represents turbulence kinetic energy and c represents local gradient Mach number reflects the compressibility of structures, and
speed of sound. Referring the data calculated by DNS, Zeman suggested referring to the correction form of expansion dissipation term based on
that[13] the ratio of ρεd and ρεs is related to Mt . The general form of turbulent Mach number:
dilatation dissipation is given below ( )
εd = F Mt , Mg εs (26)

374
C. Yi et al. Acta Astronautica 210 (2023) 372–383

With reference to the form of Sarkar and Wilcox’s correction func­

tion, the following form of the new compressible correction (will be
abbreviated as NCC below) is given, where C1 , C2 , and C3 are coefficients
which will be calibrated later in the simple shear flow:
( )
(27)
2
F Mt , Mg = C1 M 2t + C2 Mt Mg + C3 M g

2.4. Coefficients calibration

A series of experimental data on compressible turbulent mixed layers

obtained by Goebel[20] was chosen for the calibration of the co­
efficients, of which Case 4 was selected, the freestream conditions are
shown in Table 1 below. The range of the coefficients is roughly given by
experience, C1 ∈ [0, 2], C2 ∈ [− 3, 3] and C3 ∈ [0, 10] where the co­
efficients were sampled using optimal Latin hypercube sampling in three
dimensions.
Fig. 1 shows a comparison of the calculated velocity profiles with the
experimental data. The blue line indicates the calculated results of the Fig. 1. Comparison of velocity profiles between numerical and experi­
standard k-ω model. The grey area indicates the calculated results of the mental results.
samples that were taken, which essentially covers the experimental data
basic. The red line indicates the result which is in the best agreement
with the experimental data in all samples, whose coefficients are taken
as C1 = 1.16, C2 = 2.04 and C3 = 9.16, noted as k-ω NCC.
According to the combination of the velocity profiles in x-direction
and the contour of the Mach number between standard k-ω and k-ω NCC
(see Fig. 2), it is obvious that k-ω NCC corrects the spreading rate of the
mixing layer to an appropriate value, and the generalizability of this
correction will be validated in the following cases.
Through the calibration based an experimental data, the final func­
tion of compressible correction is given below:
( )
(28)
2
F Mt , Mg = 1.16M 2t + 2.04Mt Mg + 9.16M g

3. Computational details

3.1. Numerical methods

The CFD solver employed in this work is in-house code based on the
finite volume method. The code has been applied in various in­
vestigations and has displayed great performance in a range of flows
[21–24], including hear flux prediction of hypersonic vehicle[21] and
complicated shock wave/turbulence boundary layer interactions[23],
etc. In terms of numerical simulation details, the inviscid flux is dis­ Fig. 2. Contour of Mach number calculated by k-ω Standard and k-ω NCC.
cretized and reconstructed using the Roe scheme with second order
Monotone Upstream-center Conservation Law (MUSCL), while the location of reattachment. Therefore, this example is used to verify the
viscous flux is discretized using the center difference approach. To correction effect of different compressible corrections. The computa­
maintain computing stability and efficiency, the implicit Lower-Upper tional configuration and flow field structure are given by the figure as
Symmetric Gauss-Seidel (LUSGS) scheme is used in the time marching follow (see Figs. 3 and 4).
method. The free-stream conditions depend on Settle[25], etc. The
free-stream Mach number Ma∞ = 2.92, the unit Reynold number Re =
3.2. Geometry and free condition 6.7 × 107 m− 1 , the free-stream pressure P∞ = 21240Pa and the

3.2.1. Slope-cavity
In complex configuration, hypersonic boundary layer separation and
reattachment are common phenomenon. Slope-cavity flow involves
many complex flows structure like the turbulent boundary layer and free
shear layer, recirculation flow, reattached boundary layer and induced
oblique shock wave, etc. The compressible effect is often ignored in
calculations when using conventional turbulence models, which results
in the inaccurate prediction of the size of recirculation zone and the

Table 1
Freestream conditions of compressible turbulent mixed layer.
Quantity M1 , M2 T∞1 , T∞2 , K ρ∞1 , ρ∞2 , kg/m3 p∞ , kPa

Value 2.35, 0.3 171.06, 284.87 0.733, 0.44 36

Fig. 3. Slope-cavity configuration.

375
C. Yi et al. Acta Astronautica 210 (2023) 372–383

Fig. 6. The topological structure of SWTBLIs flow.

The boundary conditions of k, ω, and μt are the same as which

mentioned in Section 3.2.1.

3.3. Grid independence analysis

3.3.1. Slope-cavity
In this paper, multi-block structured grid is adopted, and for slope-
Fig. 4. The topological structure of Slope-cavity flow. cavity configuration, the depth of cavity is 25.4 mm, the length of bot­
tom is 61.9 mm, the angle of slope is 20◦ .The grid is shown in Fig. 7.
free-stream temperature T∞ = 95.37K. Simultaneously, the wall A study of three sizes of grid has been taken to ensure that the
boundary condition is altered to an adiabatic and no-slip solid wall number of grid points can satisfy the requirements of calculation accu­
racy while also reducing the demand for computer resources, and the
condition, while the values of turbulent scales k = 0 and ω = ρ60 μ
βd2
are set.
details of the different grids are listed in the Table 2. The grid Reynolds
The far-field boundary condition is specified as a uniform freestream.
number, which is used to express the grid fineness, is given by ReΔ in the
For slope-cavity, the Z planes at the side of the field are set as the general
following table. The lower the value of ReΔ , the finer the grid. The
symmetry plane. The freestream turbulence kinetic energy, specific
formula for the computation is as follows, Δ is the height of the first-
dissipation rate, and eddy viscosity are specified as k = 9 × 10− 9 , ω = layer grid:
1 × 10− 6 and μt ≈ 0.09, respectively. Where x’ is the distance in milli­
meters from the bottom of the slope to the top of the slope. Although the ReΔ =
ρ∞ u∞ Δ
(29)
freestream Mach number in this case is not very high (Ma∞ = 2.92), due μ∞
to the strong shear flow caused by the backward step, predicting the When the wall pressure and skin friction coefficients computed by
spreading rate of the mixed layer and the reattachment position of the three different grids are compared (see Fig. 8), it is discovered that the
recirculation flow remains a difficult problem, necessitating the appli­ calculated results of medium meshes and the fine meshes are very
cation of compressibility corrections. similar. Medium meshes are used in this paper to account for overall
calculation accuracy and efficiency.
3.2.2. Incident oblique shock wave/boundary layer interactions
Incident oblique shock wave/boundary layer interaction is one of the 3.3.2. Incident oblique shock wave/boundary layer interactions
typical flow structures in hypersonic inlets flow field. By changing the The grid of Incident oblique shock wave/boundary layer interactions
incident angle of oblique shock wave, different interference phenomena is shown as below (see Fig. 9).
is produced. When the incident angle of oblique shock wave is 14◦ , there The Table 3 lists the specifics of the various grids.
will be an obvious separation in the flow field. The prediction of the Below are the findings of three different fineness grids (see Fig. 10). It
location and size of separation zone will influence the accurate predic­ can be seen that both medium and fine grids can compute flow sepa­
tion of surface pressure distribution. The schlieren image obtained in the ration at the wall due to shock wave/turbulent boundary layer inter­
experiment[26] and topological structure are shown in Figs. 5 and 6 ference, and the results are identical. To calculate, the medium mesh is
below. chosen once more.
The free-stream conditions are Ma∞ = 5, total pressure P0 =
2.12Mpa, total temperature T0 = 410K, wall temperature Tw = 300K.

Fig. 5. Schlieren image obtained by experiment. Fig. 7. The grid of slope-cavity.

376
C. Yi et al. Acta Astronautica 210 (2023) 372–383

Table 2 standard k-ω model is shown in the accompanying figure. The turbulent
Details of Grid independence analysis. Mach number value surpasses 0.25 in the region near the wall where the
Grid Grid Resolution Δ ReΔ Number of grid flow reattachment is placed, which may have adverse impacts on the
Name points near wall region, except for the shear layer of free high-speed incoming
Block1 Block2
flow and separated flow in the hollow, as shown in the picture. To
G1 92 × 81 181 × 4.5 × 30 48,140 produce the final modified action region, the threshold value of turbu­
228 10− 4
G2 137 × 271 × 3.0 × 20 108,120
lent Mach number was added to Wilcox’s correction, and the near-wall
121 341 10− 4 blending function was introduced. In addition to the separation of the
G3 205 × 405 × 1.5 × 10 246,848 shear layer and the reattached boundary layer, the gradient Mach
181 509 10− 4 number at the reattached oblique shock wave was also large, reflecting
the compressibility of the structure at the shock wave, according to the
gradient Mach number distribution.
4. Numerical results and discussion
Fig. 16 depicts the contour of eddy viscosity distribution calculated
by various correction models. Different correction methods can reduce
4.1. Slope-cavity
the eddy viscosity of the model by lowering the turbulent kinetic energy
production term, which has a significant impact on the flow reattach­
In terms of dilatation dissipation, Wilcox’s and Sarkar’s models are
ment region and the boundary layer flow after the reattachment. The
applicated to k-ω model respectively. Heinz’s model was also introduced
traditional correction methods are more conservative than NCC, so the
for comparison. The computational results and experiment data are
correction effect is not as obvious as NCC.
compared in the figures below (see Figs. 11, 12 and 13), in which the k-ω
The k-ω NCC, which has the most obvious correction effect, is chosen
NCC represents the correction model of dilatation dissipation term
for comparison with the original k-ω model in this case. The addition of
raised in this paper.
the compressible correction NCC increases the size of the recirculation
We can see from the wall pressure distribution (see Fig. 11) and skin
friction coefficient (see Fig. 12) curve that, while Wilcox and Sarkar used
different modeling formulas for the dilation dissipation term, the
calculation results are relatively similar, and the results are closer to the
experimental data than Heinz’s model, which has a less obvious
correction effect. The compressible correction method NCC, which is
based on gradient Mach number Mg , achieves a better effect than the
other two correction methods for the dilatation dissipation term and is
closest to the experimental data. In comparison to other corrections, the
results show that the values of pressure plateau before and after reat­
Fig. 9. The grid of SWTBLIs.
tachment obtained by k-ω NCC are higher and more consistent with the
experimental data. The position of reattachment is delayed so as to be in
better agreement with the experimental data. However, there is no Table 3
correction for the growth rate of pressure, which is still higher than in Details of Grid independence analysis.
the experiment, i.e., the high pressure plateau after reattachment is Grid Grid Resolution Δ ReΔ Number of
reached earlier. Although the friction coefficient calculated by k-ω NCC Name grid points
Block1 Block2 Block3 Block4
is lower than the experimental data, it is still very close to other modified
values on the whole. Since the results of Wilcox’s model and Sarkar’s G1 17 × 55 × 31 × 101 × 4.5 × 30 36,872
155 143 139 227 10− 4
model are very similar, the Wilcox correction will be primarily consid­
G2 25 × 81 × 45 × 153 × 3.0 × 20 82,576
ered for comparison in the following analysis. 233 213 205 337 10− 4
The spatial distribution contour of turbulent Mach number (see G3 37 × 121 × 65 × 229 × 1.5 × 10 183,904
Fig. 14) and gradient Mach number (see Fig. 15) estimated by the 349 317 305 501 10− 4

Fig. 8. Wall Pressure and Skin Friction Coefficient of different grids.

377
C. Yi et al. Acta Astronautica 210 (2023) 372–383

Fig. 10. Wall Pressure and Skin Friction Coefficient of different grids.

Fig. 11. Wall Pressure distribution obtained by different models. Fig. 12. Skin Friction Coefficient distribution obtained by different models.

region, raises the free shear layer, and moves the reattachment position is still some discrepancy between the numerical predictions and the
backward, as shown by the Mach number contour map and streamline experimental data, the k-ω Wilcox, Sarkar, and Heinz corrections have
pattern in the flow field (see Fig. 17). As a result, the calculated pressure pretty good and reasonable progress in terms of wall pressure distribu­
and skin friction coefficient rising stages move backward, closer to the tion. Among the above corrections, Wilcox’s correction effect is the best,
experimental values. However, the vortex structure in the cavity while Heinz’s correction result is the least obvious. Although the pres­
computed in compressible correction NCC does not match that observed sure rising region after the separation zone calculated by k-ω NCC is
in the experiment, and the scale of vortex behind the step dramatically quite consistent with experimental data, the predicted separation zone’s
increases. This also indicates that changes in eddy viscosity may not starting location is too early, and the separation zone’s predicted size is
always represent the correct flow field structure. unacceptably overestimated. In the calculation of skin friction coeffi­
cient, k-ω NCC also has the problems of unreasonable calculation of the
separation zone and low peak value of skin friction coefficient after
4.2. Incident oblique shock wave/boundary layer interactions
reattachment. However, the accuracy is relatively high in the rising
stage of friction resistance coefficient after separation reattachment.
The incident oblique shock wave/boundary layer interferences are
The wall heat flux is represented by the Stanton number along the
then subjected to several compressible correction models for model
lower wall in the Fig. 20 below. This paper’s Stanton number calculation
verification and validation. The computed pressure and skin friction
formula is as follows:
coefficient distribution curves of the lower wall are shown in Figs. 18
and 19 below. The computed separation region is often too small, which q̇
St = (30)
is a common problem for the k-ω model. Different compressible cor­ ρ∞ U∞ cp (T0 − Tw )
rections can help with separation region size prediction. Although there

378
C. Yi et al. Acta Astronautica 210 (2023) 372–383

Fig. 13. Velocity profiles at different positions.

Fig. 14. Contour of turbulent mach number. Fig. 15. Contour of gradient mach number.

where q̇ is the wall heat flux, cp is the Specific heat at constant pressure, is reflected in the experimental data as well, the eddy viscosity turbu­
lence model approach calculates results where there is a valley in the
T0 is the freestream total temperature, and Tw is the local wall temper­
heat flux at the start of separation and no plateau in the separation zone.
ature. The figure shows an error occurred in the calculation of wall heat
This may be due to a flaw that caused by the application of a simple
flux via different models in the separation area, which can be seen by
Reynolds analogy in the heat flux calculation in the eddy viscosity model
comparing with the experimental data. Different from the actual phys­
itself. The correction model NCC agrees well with the experimental re
ical flow, where separation leads to a plateau in the wall heat flux, which

379
C. Yi et al. Acta Astronautica 210 (2023) 372–383

Fig. 16. Contour of eddy viscosity calculated by different models.

Fig. 17. The flow structures of traditional k-ω model and k-ω NCC.

sults in the heat flux calculation after separation. models. Different models reduce eddy viscosity in the turbulent
To characterize the action regions of different corrections, the dis­ boundary layer and separation zone near the wall, which is similar to
tribution regions of turbulent Mach number and gradient Mach number slope cavity. The correction effect of NCC is extremely visible that the
in the flow field were compared (see Figs. 21 and 22). The contour maps eddy viscosity in the separation zone and the boundary layer after
show that the regions with a higher turbulent Mach number are reattachment is considerably diminished.
concentrated at the incident location of an oblique shock wave on the In Fig. 24, we show the Mach number contour plots and streamline
lower wall, which is relatively close to the wall, whereas the regions plots calculated by the standard k-ω model and its correction model k-ω
with a higher gradient Mach number are concentrated at the incident NCC, which show that k-ω NCC predicts the boundary layer separation
shock waves and reflected shock waves rather than being confined to the location much earlier and the separation zone size increased by
high turbulence energy region near the wall. As a result, the eddy vis­ approximately five times.
cosity at the point where the shock wave passes through decreases,
resulting in the rapid induction of boundary layer separation, which 5. Conclusion
advances the separation position and increases the size of the separation
region. The feasibility of modeling the dilatation dissipation term with the
Fig. 23 shows contour maps of eddy viscosity calculated by various gradient Mach number as the detection parameter for whether it is a

380
C. Yi et al. Acta Astronautica 210 (2023) 372–383

Fig. 18. Wall Pressure distribution obtained by different models. Fig. 20. Stanton Number distribution obtained by different models.

Fig. 21. Contour of turbulent mach number.

Fig. 19. Skin Friction Coefficient distribution obtained by different models.

compressible effect zone is investigated in this work.

1. This work proposes a more general form of correction based on the k-

ω model that takes reference from traditional compressible correc­
tion methods and includes both the turbulent Mach number and the
gradient Mach number as local parameters of the flow field. The
coefficients of the new correction function are calibrated by optimal
Latin hypercube sampling method, combined with experimental data
on compressible turbulent mixing layers flow, and results are ob­ Fig. 22. Contour of gradient mach number.
tained that are in agreement with the experimental data.
2. In order to validate and verify the new proposed compressible coefficient and the wall heat flux where the separation reattached,
correction method, two cases of calculations were taken for the which demonstrated the superiority of the new approach.
slope-cavity and the incident oblique shock wave/boundary layer 3. The uncertainties in representing the Reynolds stress proportional to
interactions, and compared with the traditional compressible the mean gradient of velocity in the eddy viscosity model itself and
correction method. The new correction achieved more accurate re­ the uncertainties in the choice of the specific function form and co­
sults compared with the traditional correction approaches in the efficients calibrated by the compressible turbulent mixed layers flow
calculation of the wall pressure distribution, the skin friction only are not considered rigorously in this paper. There is still an

381
C. Yi et al. Acta Astronautica 210 (2023) 372–383

Fig. 23. Contour of eddy viscosity calculated by different models.

Fig. 24. The flow structures of traditional k-ω model and k-ω NCC.

inexcusable error in the calculation of the separation zone, which is [2] J. Smagorinsky, General circulation experiments with the primitive equations,
Mon. Weather Rev. 91 (1963) 99–164.
smaller than experimental data which is calculated by the k-ω model
[3] O. Reynolds, On the dynamical theory of incompressible viscous fluids and the
and its traditional compressible corrections, and the size of the determination of the criterion, in: Proceedings of the Royal Society of London.
separation zone obtained by the new correction is over-predicted. Series A: Mathematical and Physical Sciences, 451, 1895, pp. 5–47.
These issues still need to be further investigated for a better solution. [4] C.L. Rumsey, Application of Reynolds stress models to separated aerodynamic
flows, in: Springer Tracts in Mechanical Engineering, Springer Tracts in Mechanical
Engineering, 2015, pp. 19–37.
[5] J.A. Schaefer, V.J. Romero, S.R. Schafer, B. Leyde, C.L. Denham, Approaches for
Declaration of competing interest Quantifying Uncertainties in Computational Modeling for Aerospace Applications,
2020.
[6] M. Morkovin, Effects of compressibility on turbulent flows, in: A. Favre (Ed.),
The authors declare that they have no known competing financial Mecanique de la Turbulence, Gordon and Breach, New York, 1962, pp. 367–380.
interests or personal relationships that could have appeared to influence [7] S. Catris, B. Aupoix, Density corrections for turbulence models, Aero. Sci. Technol.
the work reported in this paper. 4 (1) (2000) 1–11.
[8] D.C. Wilcox, Turbulence Modeling for CFD, third ed., DCW Industries., La Cañada,
CA, 2006, pp. 239–254.
Acknowledgements [9] N.N. Mansour, J. Kim, P. Moin, Reynolds stress and dissipation rate budgets in
turbulent channel flow, J. Fluid Mech. 194 (1988) 15–44.
[10] F. Grasso, D. Falconi, High-speed turbulence modeling of shock-wave/boundary-
This research was supported by the National Numerical Wind Tunnel layer interaction, AIAA J. 31 (7) (1993) 1199–1206.
Project (No. NNW2019ZT1-A03) and the National Natural Science [11] David C. Wilcox, Reassessment of the scale-determining equation for advanced
Foundation of China (No. 11721202). turbulence models, AIAA J. 26 (11) (1988) 1299–1310.
[12] A. Sriram, Joseph Mathew, Improved prediction of plane transverse jets in
supersonic crossflows, AIAA J. 44 (2006) 405–408.
References [13] O. Zeman, Dilatation dissipation: the concept and application in modeling
compressible mixing layers, Phys. Fluid. Fluid Dynam. 2 (1990) 178–188.
[14] S. Sarkar, G. Erlebacher, M.Y. Hussaini, H.O. Kreiss, The analysis and modelling of
[1] S.A. Orszag, G.S. Patterson, Numerical simulation of turbulence: statistical models
dilatational terms in compressible turbulence, J. Fluid Mech. 227 (1991) 473–493.
and turbulence, the physics of dew, breath figures and dropwise condensation, The
[15] J.L. Brown, Turbulence Model Validation for Hypersonic Flows, AIAA Paper, 2002.
Physics of Dew, Breath Figures and Dropwise Condensation (1972) 127–147.

382
C. Yi et al. Acta Astronautica 210 (2023) 372–383

[16] F.R. Menter, Two-equation eddy-viscosity turbulence models for engineering [22] D. Tang, J. Li, F. Zeng, Y. Li, C. Yan, Bayesian parameter estimation of SST model
applications, AIAA J. 32 (8) (1994) 1598–1605. for shock wave-boundary layer interaction flows with different strengths, Chin. J.
[17] C. Pantano, S. Sarkar, A study of compressibility effects in the high-speed turbulent Aeronaut. 36 (4) (2023) 217–236.
shear layer using direct simulation, J. Fluid Mech. 451 (2002) 329–371. [23] H. Liu, C. Yan, Y. Zhao, Y. Qin, Uncertainty and sensitivity analysis of flow
[18] N. Wang, T. Yang, Z. Ren, Active subspace variation and modeling uncertainty in a parameters on aerodynamics of a hypersonic inlet, Acta Astronaut. 151 (2018)
supersonic flame simulation, AIAA J. 59 (5) (2021). 703–716.
[19] S. Heinz, A model for the reduction of the turbulent energy redistribution by [24] Y. Zhao, J. Chen, R. Zhao, H. Liu, Assessment and improvement of k-ω-γ model for
compressibility, Phys. Fluids 15 (2003) 3580–3583. separation-induced transition prediction, Chin. J. Aeronaut. 35 (11) (2022)
[20] S.G. Goebel, J.C. Dutton, Experimental study of compressible turbulent mixing 219–234.
layers, AIAA J. 29 (4) (1991) 538–546. [25] G.S. Settles, D.R. Williams, B.K. Baca, S.M. Bogdonoff, Reattachment of a
[21] Y. Zhao, C. Yan, X. Wang, H. Liu, W. Zhang, Uncertainty and sensitivity analysis of compressible turbulent free shear layer, AIAA J. 20 (1) (1982) 60–67.
SST turbulence model on hypersonic flow heat transfer, Int. J. Heat Mass Tran. 136 [26] E. Schülein, Skin-friction and heat flux measurements in shock/boundary layer
(2019) 808–820. interaction flows, AIAA J. 44 (8) (2006) 1732–1741.

383