RANS simulations of wind flow at the Bolund experiment

D. Cabezón1), J. Sumner2), B. Garcia1), J. Sanz Rodrigo1), C. Masson2)

1) Department of Wind Energy, National Renewable Energy Centre (CENER), Spain2) Department of Mechanical Engineering, École de Technologie Supérieure (ETS), Montreal, Canada

CASO 1: 270º CASO 2: 239º

CASO 3: 255º CASO 4: 90º

EWEC 2011, Brussels, 16th March 2011

OVERVIEW

1. Introduction

2. Bolund blind comparison

3. CFD wind flow models

1. CENER SBL model CFDWind 1.0

2. CENER ABL model CFDWind 2.0

3. ETS SBL model

4. Results1. Speed-up factors

2. Normalized turbulent kinetic energy

5. Conclusions

1. INTRODUCTIONAs part of the development of wind flow models, uncertainties must be:

Identified and evaluated

Minimized as much as possible

Method: Validation through field measurement campaigns

Since Askervein (1983), no other detailed experiment for code validation purposes carried out

Bolund blind comparison:

Extensive measurement campaign over a 12m high coastal island

Validation of wind flow models in complex terrain

Measurement of uncertainty for present state-of-the art models

Summary of 3 CFD wind flow models

2. BOLUND BLIND COMPARISONJoint project of RISOe-DTU and Vestas during 3 years (2007-2009)

Complete dataset for validation of wind flow models in complex terrain

Open call to research centres, universities and industry

Wide variety of wind flow models: Experimental methods Linearized models Non-linear CFD models: RANS (1 equation) + RANS (2 equations) + LES

Bolund hill located to the north of RISOe National Laboratory (DK)

Images property of RISOe-DTU

2. BOLUND BLIND COMPARISONSteep escarpmentComplex geometry:

12m high 75m width 130m long

Roughness change: sea (z0=0.0003m) - land (z0=0.015m)

Well-defined inflow conditionsCommon information to all the participants:

Topography and roughness description Inflow velocity and turbulence profiles

Coriolis and thermal effects neglected

Images property of RISOe-DTU

2. BOLUND BLIND COMPARISON3 months of measurements2 lines (line A and line B) with 10 meteorological masts including:

12 cups, 23 sonics, 2 LIDARs (M2 and M9 met masts)

Masts covering 4 characteristic regions: Upstream, edge, hill centre and hill wake

Reference masts for inflow conditions: M0: western wind directions M9: eastern wind direction

4 flow cases: 270º, 255º, 239º, 90º (results only shown for WD=270º)

-200 -150 -100 -50 0 50 100 150 200 250 300 350

Easting [m ]

-150

-100

-50

0

50

100

Nor

thin

g [m

]

M 0

M 1M 2

M 3

M 4

M 5

M 6M 7 M 8

M 9

Line B (270º)

3. CFD WIND FLOW MODELSCENER surface boundary layer (SBL) model CFDWind 1.0

RANS equations with turbulence closure based on kε model

Coefficients calibrated for SBL flows (Panofsky):

Cµ =0.033, C1ε=1.176, C2ε =1.92, σk=1.0,σε =1.3

Adapted for the simulation of wind speed and turbulence based on: Monin–Obukhov theory Richards and Hoxey computational approach

Turbulent viscosity computed as:

Mixing length strictly increasing with heigth: [κ~0.4]

Coriolis and thermal effects neglected

Computational model widely used by most of wind flow models based on the RANS approach with 2 equations turbulence closures

kjk

t

ii

i

Gx

k

xku

x

k

CGk

Cxx

ux k

j

t

ji

i

2

21

2kCt

lm=κz

[ε][k]

3. CFD WIND FLOW MODELSCENER atmospheric boundary layer (ABL) model CFDWind 2.0

Based on the limited-length scale kε model of Apsley and Castro Wind flow solved from the ground up to the geostrophic level 2 main differences:

Activation of Coriolis effect Limitation of mixing length lm to a maximum value lmax (avoiding too deep ABL)

Substitution of ε production term at the ε equation by:

where: , [Ug=geostrofic wind, f=Coriolis factor]

Mixing length affecting turbulent viscosity and turbulent transport

Coefficients calibrated for ABL flows (Detering and Etling):

Cµ =0.0256, C1ε=1.13, C2ε =1.9, σk=0.74,σε =1.3

k

CGk

Cxx

ux k

j

t

ji

i

2

21

k

Gl

lCCC k

m

max121 )(

2/34/3 kC

lm

c

g

f

Ul

00027.0max

ETS surface boundary layer (SBL) model RANS equations with turbulence closure based on the RNG kε model Additional ε source term by:

where η = RNG coefficient, function of the magnitude of the mean strain tensor S β, η0 = constant RNG coefficients

Improvement of flow predictions where recirculation is present

Coriolis and thermal effects neglected

Coefficients calibrated for SBL flows (El Kasmi and Masson):

Cµ =0.0333, C1ε=0.47, C2ε =1.68, σk=1.0,σε =1.3, β=0.012, η0 =4.38

3. CFD WIND FLOW MODELS

k

CGk

Cxx

ux k

j

t

ji

i

2

21

k

CP

2

3

03

1

)/1(

Computational features of the models for the simulations at Bolund

3. CFD WIND FLOW MODELS

Model Domain Extents

Grid Boundary conditions Numerical method

Inlet / outlet

Wall Sides / top

CFDWind1.0 (SBL)

1260m (E-W)1170m (N-S)300m (vertical)

ICEM CFD3M cells∆x=∆y=0.8mZ1=0.025m

MO profiles / zero pressure

Modified wall funct. (Blocken et.al)

Symmetry /fixed shear stress

FLUENT 12.0Second-order upwind

CFDWind2.0 (ABL)

1260m (E-W)1170m (N-S)1700m (vertical)

ICEM CFD4.25M cells∆x=∆y=0.8mZ1=0.025m

1D ABL profiles / zero pressure

Periodic / symmetry

ETS (SBL) 990m (X-axial)650m (Y-cross)120m (Z-vertical)

BlockMesh8M cells∆x=0.5m∆y=2mZ1<=0.1m

MO profiles / zero gradient

z0-based wall functions

Symmetry /fixed shear stress

OPENFOAM 1.5.x

Central-differencing / First-order upwind

4. RESULTSSpeed-up factor (U/Uref)

Results only for 270º inflow direction

Along line B at 2m and 5m high

Fairly good agreement with predictions

Influence of RNG downstream the first escarpment

Influence of lm limiting effect in the wake of the hill

Z (m) CFDWind1 CFDWind2 ETSave 0.087 0.109 0.126min 0.001 0.001 0.034max 0.313 0.297 0.196ave 0.051 0.058 0.100min 0.021 0.001 0.003max 0.081 0.184 0.186

2m

5m

Z=2m agl Z=5m agl

4. RESULTSNormalized turbulent kinetic energy (k/U2

ref )

Results only for 270º inflow direction

Along line B at 2m and 5m high

Overestimation (except first edge -2m high)

One single zone of elevated k from experiments

Biggest overestimation produced by CFDWind1.0

Z (m) CFDWind1 CFDWind2 ETSave 0.024 0.026 0.024min 0.001 0.001 0.006max 0.048 0.059 0.060ave 0.011 0.008 0.010min 0.001 0.000 0.000max 0.036 0.026 0.027

2m

5m

4. RESULTSTurbulence length scale lm

Axial evolution for 270º inflow direction (line B at 5m high) Related to turbulent viscosity according to:

5.025.0 kC

l tm

Downstream of the first escarpment: peak in k leading to peak in lm

In the wake of the hill, CFDWind2 shows:

Quicker reduction of ε near the wall

Rapid increase of turbulent viscosity and faster flow recovery

Even that, further investigation required

4. CONCLUSIONSCFD wind flow models based on RANS approach and 2-equation kε closure:

State-of-the-art in wind flow modeling

Properly modified to represent SBL / ABL

Accurate predictions of mean wind speed [mae(speed-up factors ~ 10-1)]

Further improvement needed in turbulence modeling [mae(normalized k ~ 10-2)]

Full ABL models with respect to SBL models

Improve predictions for big hub heigths outside the SBL

Consequences of the limiting-length–scale effect near the ground to be investigated

Further work and research needs:

Improve turbulence modelling as much as possible (RSM, LES, DES, etc.)

Mesoscale-microscale coupling (generation of boundary conditions, high resolution wind maps, etc.) -> european wind atlas!

Validate at new sites based on massive measurement campaigns

Create collaborative research networks

…

ACKNOWLEDGEMENTS

We would like to acknowledge A.Bechmann, P.E.Rethore, N.N. Sørensen, J. Berg,

H.E. Jørgensen, J. Mann, M. Courtney, P. Hansen and the rest of the team at

RISOe-DTU and Vestas for organizing and funding the Bolund blind

comparison and supplying the database for the validation of models

Image property of RISOe-DTU