You are on page 1of 11

Improving the modelling of Laminar Separation

Bubbles

Louis GAUVAIN / MNA1

February 2, 2023

Professor : Yannick HOARAU / Abderahmane MAROUF

1
Contents

1 Introduction 3

2 Methodologie 6
2.1 RANS equations [3] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 γ − Reθ Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 NSMB Solver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2
Chapter 1

Introduction

The use of Computational Fluid Dynamics (CFD) is gaining a bigger advantage nowadays
in the fields of aerodynamics and marines. But most of those applications are based on the
ideas that those object evolve in a turbulent field. This might be true in a lot of situation,
for big or fast moving objects for exemple but there is also a lot of situation where this is not
true and where we therefore need to take in account the transition phase between turbulent
and non-turbulent (laminar) flow. This is the kind of applications that the Reynolds-Averaged
Navier-Stokes equations (RANS) is unable to predict. It can be used to predict the location
of these transition phase but not characterise them.

There exist multiple type of transition [2] :


Natural : Transition that emanate from exponential instabilities such as T-S Waves or cross-
flow.

Figure 1.1: Schematic representation of Natural transition

Bypass : When the induced flow passes from a Laminar Regime to a Turbulent one through
some secondary instability mode, Bypassing some of the pre-transitional events that
typically occur in a natural laminar-turbulent transition.

3
Crossflow : They appear when a new velocity components develop near the wall in the sweep
direction on a wing.

Separation Induced : When the boundary layer detaches under the influence of an adverse
pressure gradiant and the separated shear layer undergoes transition because of the dis-
turbances amplification in the unstable laminar layer. The momentum transfer in the
normal direction, caused by the turbulent mixing, eventually eliminates the reversed
velocities near the wall causing the boundary layer to reattach. This recirculation zone
is called Laminar Separation Bubble (LSB), whose schematic representation is shown in
Figure 2 The zone bounded by the surface and the dividing streamline ST ′ R represents
the re-circulatory flow forming the bubble. The zone between the divided streamline
and the outer edge of the boundary layer is the separated shear layer, where the flow
undergoes transition at the location T . R is the reattachment location The effect of
the laminar separation bubble on the overall pressure coefficient distribution is shown
in Figure 2. As the flow separates at location S the edge of the bubble ahead of the
transition position, T , is a zero pressure gradient streamline, as denoted by the pressure
plateau.

The primary instability in a separation bubble is inflectional. It originates upstream of


the separation point, because of the strong adverse-pressure gradient, and it is convected
downstream. The laminar separation bubble length depends on Reynolds number and
free-stream turbulence intensity. As T u and Re increase, the bubble length is reduced
and the suction peak increases in magnitude. On the other hand, for decreasing Re
the viscous damping effects become significant, suppressing transition and, eventually,
delaying reattachment. The angle of attack α plays a major role on the separation
point, increasing α the separation point moves upwards, because of the stronger pres-
sure gradient at higher incidences. The distinction between short and long bubbles
depends on their effect on the overall pressure distribution. A short separation bubble
affects the pressure distribution locally, because it reattaches shortly after the separation
point. Nevertheless, a short separation bubble might "burst" into a long one or into an
unattached shear layer, causing a considerable loss of lift.

Figure 1.2: Schematic representation of Laminar Separation Bubble

4
The first model that was used to determine Laminar to turbulent transition was the LES
(Large Eddy Simulations), it was widely used until the 21th century. This method unfortu-
nately demand high computational ressource to work. Another problem is that LES related
are sensitive at a subgrid scale. To avoid this problem, the Wall-Resolved-LES(WRLES)
method was developed but it is not already usable. Seeing the difficulty of developing such
kind of method, many researchers decided to focus their effort on RANS models able to handle
automatically and autonomously transition. The main implication with RANS is that the av-
eraging eliminates the linear amplification of the disturbances and the discarded linear effects
might seem not compatible with transition physics. Nevertheless, one has to consider that, in
many applications, transition occurs in small flow area and it is dictated by geometry features,
pressure gradients, and flow separations. A RANS model is able to capture these effects with
sufficient engineering accuracy, upon the inclusion of proper correlations in the models formu-
lation. Local correlation-based transition models (LCTM) have been proposed based on this
idea. The equations do not intend to model the real physics, the aim is to identify the region
of the flow that is laminar and the transition location. The most succesful is the eN method
of Smith Gamberoni and Van Ingen.

There exist many RANS model for transition modelling. the most common model is usu-
ally the γ − Reθ model, although it does not operate alone. It needs to be coupled with a
turbulence modelling model such as the most commonly used k −ω model. Both. these model,
the γ − Reθ et the k − ω model where developed by Menter in the early 2000’s. The γ − Reθ
model coupled with the k − ω model for turbulence modelisation is only a two-dimensional
model. It is therefore only able to determine the position of 2D transition mechanism like
transition due to Tollmien-Schlichting (TS) instability, separation-induced and by-pass transi-
tion. For simulation of turbulence on an airfoil, there is often strong adverse pressure gradient.
This often result in crossflow transition that can’t be predicted by γ − Reθ models. To take
in accounts the possible crossflow transition we can improve the γ − Reθ model by adding
equations specific to this type of transition. This results in hybrid models that can be referred
to as γ −Reθ −CF models. We can also try to couple the γ −Reθ model with the SSG−LRRω
model for turbulence modelling that dispose of an increased order of turbulence modelling. [1]

There also exist variant of the γ − Reθ model such as the γ model which is a simplified
version of the previous. It is better to implement in complexe code because it is much simpler.
It is also able to calculate separation for moving walls unlike the γ − Reθ model that can only
be used for static wall computation.

5
Chapter 2

Methodologie

2.1 RANS equations [3]


Many, if not all, industrial codes for CFD use RANS equations. Their goal is to represent
the action of turbulence on the average flow through the least expensive statistical approach.
In the case of turbulence modelling, the most used form of these equations is time averaging.
The time average of a quantity φ(x, t) is
󰁝 t+T
1
φ = lim φ(x, t)dt
T →∞ T t

Now that it as been averaged, the quantity φ does not depend on time. Once an averaging
operator as been defined, φ(x, t) can be re-written as

φ(x, t) = φ + φ′ (x, t)
Where φ′ (x, t) represents the fluctuating part.
With this operator, The resulting averaged Navier-Stokes equations for each component
ui are:

∂ui
=0
∂xi
󰀗 󰀕 󰀖󰀘
∂ui ∂ui 1 ∂p ∂ ∂ui ∂uj ∂ 󰀓 ′ ′󰀔
+ uj =− + ν + + −ui uj
∂t ∂xj ρ ∂xi ∂xj ∂xj ∂xi ∂xj
An unclosed term appears from the averaging, the Reynolds stress tensor, whose compo-
nents are τij = −u′i u′j . This tensor quantifies the influence of the velocity fluctuations over
the mean flow. The primary problem with RANS is that there are more unknowns (pressure,
three velocity components, six components of the symmetric tensor τij ) than equations and
additional relations are needed to close the system.

6
2.2 γ − Reθ Model
The main model that is used to predict Laminar to turbulent transition is te γ − Reθ model.
It is par tof the LCTM models and was first proposed by Menter et al in 2006 [5]. It is coupled
with the k − ω model for turbulence modelling [4].

The Reθ parameter stands for the momentum thickness Reynolds number. It is calculated
from Reν and is equal .

Reνmax
Reθ ≈ (2.1)
2.193
Where
ρy 2
Reνmax = S
µ
Where y is the distance to the nearest wall, ρ is the density, µ is the dynamic viscosity
and S is the absolute value of the strain rate.

Figure 2.1: Scaled vorticity Reynolds number on a Blasius boundary layer

When the pressure gradient become too strong, the relationship between momentum thick-
ness and vorticity Reynolds number described by equation (1) changes due to the change in
the shape of the profile.

The γ stands for the intermittency. It is a characteristic of transition flows. It is defined


as the irregular alternation of phases of apparently periodic and chaotic dynamics. It can take
values between 0 and 1, 0 being a fully laminar flow and 1 being a fully turbulent flow. The
transport equation for γ is :
󰀗󰀕 󰀖 󰀘
∂(ργ) ∂(ρUj γ) ∂ µt ∂γ
+ = Pγ − E γ + µ+
∂t ∂xj ∂xj σf ∂xj
Where Pγ is the transition source defined as :

7
Pγ = Flength ca1 ρS [γFonset ]0.5 (1 − ce1 γ)
Where S is the strain-rate magnitude Flentgth is an empirical correlation that controls the
length of the transition region it is determined in functions of Re˜θt (valeurs en annexe), and
Fonset controls the transition onset location. Both are dimensionless functions that are used
to control the intermittency equation in the boundary layer.
The destruction/relaminarization source is defined as follows:

Eγ = ca2 ρΩγFturb (ce2 γ − 1)


Where Ω is the vorticity magnitude. It characterizes the length of the bubble. We also
define the following functions
󰀻
󰁁
󰁁Fonset,1 = Reν
󰁁
󰁁
󰁁
󰁁 2.193Re
󰀃 θc󰀃 󰀄 󰀄
󰁁
󰁁 4
󰁁 F
󰀿 onset,2 = min max Fonset,1 , Fonset,1 ,2
ρk
󰁁 RT =
󰁁
󰁁 µω 󰀣
󰁁
󰁁 󰀕 󰀖3 󰀤
󰁁
󰁁 R T
󰁁
󰁁Fonset,3 = max 1 − ,0
󰀽 2.5

Fonset = max (Fonset,2 − Fonset,3 , 0)


Where Reθc is the critical Reynolds number where the intermittency first starts to increase
in the boundary layer.

The constants for the intermittency equations are given by :

ce1 = 1; ca1 = 2; cα = 0.5; ce2 = 50; ca2 = 0.06;


The modification for separation induced transitions is :
󰀕 󰀗󰀕 󰀖 󰀘 󰀖
Reν
γsep = min 2 max − 1 Freattach , 2 Fθt (2.2)
3.235Reθc
󰀣 󰀤4
RT

Freattach = e 20

γef f = max(γ, γsep )


We can use the same method to determine Reθ .

The models constant in equation 2 have been adjusted from those of Menter et al in order
to improve the predictions of separated flow transition. The main difference is the constant
that controls the relation between Reν and Reθc was changed from 2.193 (Blasius Boundary
layer) to this of the value at a separation point where the shape factor the one you need to be.

8
2.3 NSMB Solver

Initially developed in 1992 by the Swiss Federal Institut of Technologie in Lausanne, the
Navier Stokes Multi Block solver (NSMB) is a CFD equations solver based on the Navier
Stokes equations. Today, it is developed in collaboration by IMF-Toulouse, the ICUBE lab-
oratory, the university of Munchen, the university of the Army in Munchen, Airbus-Safran,
RUAG Aviation and CFS Engineering. NSMB is a parallelized solver that discretize the Navier
Stokes equations by employing the cell-center finite volume method using multi block struc-
tured grids. It uses sliding mesh and the chimera method (also used for simulations involving
moving bodies) to simplify the mesh generation for complex geometries.

Turbulence can be modelled by using the k − ω model and transition to turbulence can be
modelled by solving the γ − Reθ transport equations as we’ve seen before and by specifying
transition lines or planes.

9
󰀻 󰀻
󰁁
󰁁 398.189 × 10−1 + (−119.270 × 10−4 )Re˜θt + (−132.567 × 10−6 )Re˜2θt 󰁁
󰁁 Re˜θt < 400
󰁁
󰀿 󰁁
󰀿400 󰃖 Re˜ < 596
263.404 + (−123.939 × 10−2 )Re˜θt + (−194.548 × 10−5 )Re˜2θt + (−101.695 × 10−8 )Re˜3θt θt
Flength =
󰁁
󰁁
󰁁 0.5 − (Re˜θt − 596) × 3 × 10−4 󰁁596 󰃖 Re˜θt < 1200
󰁁
󰁁
󰀽 󰀽
0.3188 1200 󰃖 Re˜θt

10
Bibliography

[1] A. Krumbein C. Grabe. Correlation-based transition transport modeling for three-dimensional aerodynamic config-
urations. Journal of Aircraft, 50(5):1533–1539, 2013.

[2] M.S. Genc. Low reynolds number aerodynamics and transition. Intechopen, 2012.

[3] G.Rubino. Laminar-to-turbulence transition modeling of incompressible flows in a rans framework for 2d and 3d
configurations, 2022.

[4] F. Menter. Two-equation eddy-viscosity turbulence models for engineering applications. AIAA Journal, 32:1598–
1605, 1994.

[5] F. Menter. A correlation-based transition model using local variables - part i: Model formulation. pages 413–422,
2006.

11

You might also like