Beruflich Dokumente
Kultur Dokumente
Atmospheric Turbulence
Sk =CD βp U 3
x − xD
− CD βd U k G (3)
D
and
ε
Sε =CD Cεp βp U 3 Figure 1: Actuator disc & mesh (1 out of 3 cells).
k
x − xD
− CD Cεd βd U ε G , (4)
D 2.4 Mesh and Boundary conditions
where p indicates a parameter that accounts
The domain is a rectangular cuboid (1200 m
for the production of small-scale turbulence, d
× 600 m × 200 m ). The mesh is horizontally
for the dissipation of large-scale turbulence, x
stretched in order to obtain a regular square
is the vector of the considered position, xD
meshing in a region surrounding the wind tur-
is the vector position of the rotor, and G is a
bine and its wake in order to have 10 cells
smoothing function that accounts for the action
per rotor diameter (∆xi < D/10). In the out-
of the pressure gradient distribution.
flow region, the mesh is stretched out in or-
der to numerically reduce the gradients and en-
Realizability model hance the convergence (Fig.1). It is vertically
stretched in order to obtain a first cell equiva-
We define the Realizability model as follow.
lent to the surface roughness length z0 at the
The concept of realizability of a Reynolds-
wall boundary condition (BC). The mesh is de-
averaged turbulence model is directly de-
composed in blocks of 643 cells. The whole
rived from the relation between the cross-
domain has the dimension of 7 × 5 × 2 blocks
correlations and auto-correlation of the velocity
(448 × 320 × 128 > 18.35M cells).
fluctuations [17]. The realizability of a flow can
The inflow BC are defined as neutrally strat-
be checked using two relations. The Schwartz
ified (i.e. following the log-law). The side BC
inequality imposes a positive normal stress (5)
are set to enforce symmetry. The LES top BC
and the inequality relationship between the off-
is also symmetric, while the k -ε top BC is set
diagonal terms and the diagonal terms (6):
as inlet. The two simulation have therefore not
Rij > 0, when i = j, (5) exactly the same inflow condition. In particu-
lar, the turbulence kinetic energy level at hub
|Rαβ |2 ≤ Rαα Rββ . (6) height is lower in the LES compared to the k -
ε computation. It would, of course, have been
The eddy-viscosity limiter νt,Swz we propose
more appropriate to have exactly the same in-
here is calculated by finding the maximum
flow condition in both cases, this was unfortu-
value that matches the Schwartz inequalities
nately not possible during the time-frame of the
based on the eddy-viscosity expression of the
project. While it is difficult to compare directly
Reynolds-stresses (6). This inequality has two
the turbulence values together, we nonethe-
solutions, where the maximum value is taken,
less believe that we can draw some conclu-
νt,Swz = max(νt,Swz1 , νt,Swz2 ). (7) sions based on the difference in the trends ob-
served in the Results Section.
In the case where both solutions are nega-
The time step is set to ∆t = 0.025 s. The
tive, there are no solution and the normal defi-
solving time of the LES for 9000 iterations (i.e.
nition for the eddy-viscosity is used.
225 s) is about 4 days on 70 CPUs. The
To extend this method, a factor on the
steady-state k -ε simulation takes only a cou-
Schwartz eddy-viscosity can be applied:
ple of hours to converge to acceptable residu-
k2 als (< 10−6 ).
νet = min Cµ , CSwz νt,Swz , (8)
ε
which gives more control over the region where
2.5 LES Turbulence Scalars
the flow is considered unrealizable. Note that
the model only has a physical basis for the con- In order to compare the steady-state k -ε with
stant CSwz = 1.0. the unsteady LES in term of turbulence, we in-
troduce here different methods to estimate the 3 Results
Resolved LES eddy-viscosity and Resolved
Reynolds-stresses. 3.1 Model Comparison
Fig.2 presents a centerline extraction passing
Resolved LES eddy-viscosities
through the rotor at hub height and at the ra-
Two definitions are used to estimate the re- dial position y = −0.25D. Three different mod-
solved LES eddy-viscosity. The first one, els are presented, the El Kasmi-Masson Model
νtLES1 , is based on the k -ε definition, (a), the Realizability Model (b) and the Wind
Turbine Canopy Model (c). For each models
k2 three different graphs present the axial velocity
νtLES1 = CµLES . (9)
ε normalized by the inflow velocity at hub height
where CµLES = 0.03 is the eddy-viscosity con- U /U ∞,H , the TKE normalized by the friction
stant used for the LES, k is the resolved LES velocity squared k/u∗2 and the dissipation nor-
TKE and ε is the resolved LES dissipation (k malized by the inflow dissipation at hub height
and ε are both found by time series averaging). ε/ε∞,H . In each graphs, different parameters
of the models are compared with the standard
The second one, νtLES2 , is based on k -ε and the unsteady force LES results.
the eddy-viscosity concept, using the time-
averaged resolved Reynolds-stresses tensor 3.2 Normal Reynolds-stress
u′i u′j directly estimated from the computation
and the mean strain rate tensor S ij . Fig.3 compares the normal axial Reynolds-
stress of k -ε and LES, which are estimated us-
1 ||R|| ing two different methods (14) and (15). The
νtLES2 = , (10)
2 ||S|| extraction line passes through hub height (y =
−0.25D).
where ||R|| and ||S|| are defined as:
s 2 LES: R11
2 60
||R|| = kδij − u′i u′j , (11) LES: u′ u′
3
50
q k-ε: R11
R11 /u∗2 [-]
40
||S|| = S ij S ij . (12)
30
Based on these two definitions, we define
the eddy-viscosity factor Cµ∗ , 20
10
||R|| ε
Cµ∗ = . (13) 0
2||S|| k 2 0 5 10
x-direction x/D [-]
According to the eddy-viscosity concept and
the k -ε assumptions, this factor should be con-
stant. This new definition is therefore used in Figure 3: Normal axial Reynolds-stress compari-
the Discussion Section to assess the regions son.
where the eddy-viscosity assumptions are not
applicable.
3.3 Turbulence Kinetic Energy
Resolved LES Reynolds-stresses
Fig.4 compares the TKE from k -ε with the Re-
We define the Reynolds-stresses either esti- solved TKE of the two LES, with steady and
mated directly by time-averaging of the square unsteady forces. The extraction line passes
of the velocity fluctuations, through hub height (y = −0.25D).
3 0.1
2 0.08
1
0.06 Most of the Reynolds Averaged Navier-Stokes
0
−1
0.04 (RANS) turbulence models rely on the eddy-
−2 0.02
viscosity concept, also known as Boussinesq’s
−3 0
−4 −2 0 2 4 6 8
x-direction x/D [-]
10 12 14 approximation. The idea is based on the turbu-
lent transport hypothesis (Fig.6).
xp2
x
4 Discussion
xp3
4.1 Increased dissipation
xp4
Some wake measurement campaigns [7, 19]
have observed that in the close wake region, Figure 6: Turbulence transport hypothesis.
the large-scales of the TKE are reduced com-
pared to the inflow conditions. Following these
Bernard and Wallace [2] identified three
observations, some papers [6, 11] have postu-
major assumptions necessary to obtain the
lated that the dissipation ε-equation is unphys-
Reynolds-stresses of the eddy-viscosity con-
ical in the region surrounding the wind turbine.
cept.
They assumed therefore that the wind turbine
somehow causes an imbalance in the produc- 1. For each flow resolution, the particle at the
tion/dissipation ratio that should be accounted position x, time t is assumed to conserve
for in the dissipation equation by increasing the its velocity during a time τ (i.e. the turbu-
dissipation locally proportionally to the produc- lent time-scale). So that
tion of turbulence.
The LES results presented in Fig.4 show Ui (xp , t − τ ) = Ui (x, t), (16)
that, in agreement with the measurement ob-
servations, the TKE is reduced in the close where xp is the position of the particle at
wake region when the forces are correlated the time t − τ .
with the local velocity (i.e. in the case of the
2. This time τ is large enough so that the
Unsteady Forces LES). However, Fig.2 shows
fluctuation components at the time t − τ
that the dissipation is decreased instead of be-
are uncorrelated with the other fluctuation
ing increased as it was assumed in [6, 11]. The
components at the time t.
dissipation in the El Kasmi-Masson model in-
creases sharply upstream of the wind turbine u′i (xp , t − τ )u′j (x, t) = 0 (17)
in disagreement with the LES results.
According to the wind turbine canopy model,
3. The mean velocity between x and xp can
there should be both a source term and a sink
be assumed to be linear.
term in k - and ε-equation. The source term
would account for the production of small-scale ∂Ui
vortex structure caused by the geometry of the Ui (xp ) = Ui (x) − Lp,j (x), (18)
∂xj
where Lp,j is the distance that the particle
went through during the time τ (i.e. the
turbulent length-scale).
x x
p
In order to define the eddy-viscosity concept,
the expression of the fluctuation component is
first expressed as
U(x ) U
p
u′i (x, t) = Ui (x, t) − Ui (x). (19) U(x)
14
L∂U/∂x
parameter found by calibrating at a given wind
12
xp condition, or even wind turbine size, would
10 be applicable in another setup. Moreover, the
U(xp)−U(x)
x region of operation of the model extends far
8
11
L∂U/∂x with more or less success. For airfoil compu-
U(xp)−U(x)
10 x tation, [8] suggests to express the turbulent ki-
9
netic energy production in terms of the vorticity
instead of the strain-rate,
8 L
7
Pk = 2νt Rij Rij , (25)
−100 −80 −60 −40 −20 0 20 40 60 80 100
y−direction [m]
k/u∗2 [-]
30
20
10
0
50
ε/ε∞,H [-]
40
30
20
10
0
2
22 0 2 4 6 8 10 12
x-direction x/D [-]
(a) El Kasmi-Masson Model
CSwz = 1.0
1
U/U ∞,H [-]
CSwz = 0.5
0.8 CSwz = 0.4
Std k -ε
0.6
LES
0.4
40
k/u∗2 [-]
30
20
10
0
30
ε/ε∞,H [-]
20
10
0
2
22 0 2 4 6 8 10 12
x-direction x/D [-]
(b) Realizability Model
βd = 1, Cεd = 1
1
U/U ∞,H [-]
βd = 1, Cεd = 0
0.8 βd = 2, Cεd = 0
Std k -ε
0.6
LES
0.4
40
k/u∗2 [-]
30
20
10
0
50
ε/ε∞,H [-]
40
30
20
10
0
2
22 0 2 4 6 8 10 12
x-direction x/D [-]
(c) Wind Turbine Canopy Model
Figure 2: Comparison of the different modification models with the standard k-ε and the LES.