Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52

Contents lists available at ScienceDirect

Journal of Wind Engineering & Industrial Aerodynamics

journal homepage: www.elsevier.com/locate/jweia

Effects of blade design on ice accretion for horizontal axis wind turbines
G.M. Ibrahim, K. Pope *, Y.S. Muzychka
Department of Mechanical Engineering, Faculty of Engineering and Applied Science, Memorial University of Newfoundland, St. John's, NL, A1B 3X5, Canada

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

Keywords: This paper presents a numerical study to predict ice loads on a wind turbine blade section at 80% of blade span.
Wind power Air flow, droplet impingement and ice accretion simulations are conducted in FENSAP ICE. The effect of low and
Atmospheric icing high liquid water content (LWC) conditions on blade thickness is presented. All NREL airfoil families used for
Numerical predictions HAWTs are examined to study blade design's effect on ice accretion. Ice shapes are numerically predicted at
Aerodynamics different atmospheric temperatures and LWC conditions. The degradation in aerodynamic characteristics due to
ice formation is predicted. The numerical predictions suggest a 10%–65% reduction in lift coefficient due to ice
accretion, which is highly dependent on the accreted ice shape.

1. Introduction the atmospheric temperature range. Additionally, numerous different ice

The installed capacity of wind turbines has expanded rapidly in recent
years. However, current major issues for further development of HAWTs,
such as fatigue problems associated with upscaling of turbines (Rezaeiha
et al., 2017), floating offshore installations (Borg et al., 2014; Paulsen
et al., 2014) and operation in harsh northern locations should be
addressed to facilitate continued rapid expansion of installed capacity.
Cold regions typically have abundant wind power resources making them
attractive locations for wind farm installations (Reid et al., 2013).
However, wind turbines installed in cold climates can experience blade
icing causing significant power losses (Hochart et al., 2008), and atmo-
spheric icing is a common cause of ice accretion on a wind turbine (Reid
et al., 2013). Ice accretion on wind turbines can have significant impacts
on its operation (i.e. aerodynamic performance and controlling systems),
safety (i.e. site accessibility due to ice throwing) and economics (i.e.,
annual energy production and turbine lifespan) (Virk et al., 2012). At-
mospheric icing on wind turbines occurs when super-cooled water
droplets collide with the surface of the blades and freeze upon impact
(Reid et al., 2013). Formed ice type is based on the atmospheric and
operational conditions, including temperature, pressure, wind velocity,
liquid water content (LWC), median volume diameter of droplets (MVD),
rotor blade curvature, roughness, heat flux, icing event time, and
collection efficiency of droplets (Battisti, 2015; Kraj and Bibeau, 2010).
Typically, ice accumulates on the turbine blades at temperatures be-
tween 0
C and 15
C (Virk et al., 2012; Battisti, 2015). Ice accretion
mechanisms are usually classified into two different types: (i) glaze ice
and (ii) rime ice (Makkonen, 2000), which are typically formed based on
shapes can be formed, however, shape prediction is difficult due to the
rapid changes in the atmospheric conditions during icing events (Fikke
et al., 2006). Glaze ice usually forms transparent, horn shapes of ice and
is typically caused by freezing rain/drizzle or wet icing conditions with
high LWC concentrations Reid et al., 2013; Virk et al., 2012; Battisti,
2015). It is formed at temperature ranges between 0
C and 5
C and
droplet diameter ranges from 0 μm to 500 μm. Its density is higher than
rime ice, usually around 900 kg/m3 (Reid et al., 2013; Virk et al., 2012;
Cattin, 2013).
Rime ice is caused by frozen fog with smaller range of droplet sizes
compared to glaze ice, ranging from 0 μm to 10 μm (Reid et al., 2013;
Battisti, 2015). Rime ice usually occurs at lower temperatures below
6
C with lower concentrations of LWC, as the temperature is well
below the freezing point water droplets freezes instantaneously upon
collision with object surface. Rime ice usually forms grainy and opaque
ice shapes (Reid et al., 2013). Depending on droplet size, soft or hard
rime can be formed. Typically, rime ice density ranges from 100 kg/m3 to
600 kg/m3 (Reid et al., 2013; Virk et al., 2012; Cattin, 2013). Atmo-
spheric measurements indicated that LWC usually varies from 0 g/m3 to
5 g/m3. Low LWC concentrations below 1 kg/m3 can be existed during
stratiform clouds environment. In contrast to cumuliform clouds envi-
ronment which are characterized by higher LWC concentrations usually
higher than 1 kg/m3 (Cober et al., 2001; Masters, 1985). LWC value is
highly influenced by elevation and atmospheric temperature. The rela-
tionship between LWC, MVD and temperature was experimentally eval-
uated showing that LWC decreases as the droplet size increases (Jeck,
2002). LWC value decreases with altitude and temperature.

* Corresponding author.
E-mail addresses: gmi588@mun.ca (G.M. Ibrahim), kpope@mun.ca (K. Pope), yurim@mun.ca (Y.S. Muzychka).

https://doi.org/10.1016/j.jweia.2017.11.024
Received 28 June 2017; Received in revised form 23 November 2017; Accepted 25 November 2017
Available online 11 December 2017
0167-6105/© 2017 Elsevier Ltd. All rights reserved.
G.M. Ibrahim et al. Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52

Nomenclature n Surface normal vector

XOffset Offset distance (m)
ρ Density (kg/m3) TMax Maximum thickness (m)
V Velocity (m/s) XMean Average distance (m)
T Temperature (K)
m Mass flow (kg) Subscripts
d Droplet diameter (μm) a Air
A Surface area (m2) d Droplet
p Static pressure (Pa) f Fluid
μ Dynamic viscosity (kg/m⋅s) evap Evaporation
E Internal energy (J) s Solid
H Enthalpy (J/kg) ∞ Reference
α Local water volume fraction (kg/m3) Rec Recovery
β Collection efficiency h Water film height
C Chord length (m) l Lift
C Coefficient d Drag
F Force (N) U Upper
K Droplet inertia parameter L Lower
τij Shear stress tensor Abbreviations
g Gravity vector (m/s2) LWC Liquid water content
Fr Froude number MVD (Droplet) Median volumetric diameter
Re Reynolds number CFD Computational fluid dynamics
L Latent heat (J/kg) HAWTs Horizontal axis wind turbines
Q Heat flux (W/m2) NREL National Renewable Energy Laboratory
σ Stefan-Boltzmann constant (σ ¼ 5.67  108 W/m2⋅K4) AoA Angle of attack
ε Solid emissivity

The most common solvers applied to investigate icing on wind tur-
bines are LEWICE (Ruff and Berkowitz, 1990), TURBICE (Marjaniemi
et al., 2000; Makkonen et al., 2001) and FENSAP ICE (H
e et al., 2003;
Habashi et al., 2001). LEWICE is 2-D an ice accretion code developed by
NASA Glenn Research Center. The code applies a time-stepping proced-
ure code to predict ice accretion shape. The flow field calculations are
computed using Douglas Hess-Smith 2-D panel code. Then the obtained
solution is used to calculate the trajectories of particles and the
impingement points on the body. The icing model is then used to predict
the ice growth rate and shape (Ruff and Berkowitz, 1990). Turbine Blade
Icing Model (TURBICE) is a two-dimensional ice accretion simulation
program. It was developed to predict icing on wind turbine blades.
TURBICE uses panel methods to calculate potential flow field around the
blade. The program uses Lagrangian technique for droplet trajectories
and impingements calculations. The program is capable of calculating
collision efficiencies and locating the droplet impingement locations
along the blade surface. The program can estimate the amount of energy
required for heating to prevent ice accretion on blade surface (Ruff and
Berkowitz, 1990; Makkonen et al., 2001). FENSAP ICE is a 3-D premier
in-flight icing simulation solver. The solver simulates flow field, droplet
impingements, ice shapes and predicting anti/de-icing heat loads using a
built in CFD modules. It uses 3D Navier-Stokes and energy equations for
flow field calculations and 3-D Eulerian model for droplet calculations
(H
e et al., 2003; Habashi et al., 2001).
Ice accretion on wind turbines can greatly degrade the aerodynamic
performance and reduce the power production by up to 60% at lower
wind speeds (Reid et al., 2013). In a study on icing of large wind turbines,
the results indicated that ice growth can be greatly influenced by com-
bined changes in blade design (i.e., size and profile) and relative velocity
at each blade section (Virk et al., 2012).
Furthermore, ice accretion was observed to be more at blade sections
near the tip compared to root section, where both blade thickness and
chord length are minimized (Virk et al., 2012). A study by Homola (Virk
et al., 2010a) indicated that ice growth in terms of ice rate and accretion
shape is highly influenced by angle of attack of fluid flow. A study of icing
on small wind turbines identified that blade rotation can lead to large ice
accumulation at blade sections near the tip and at the leading edge of a
blade section (Bose, 1992a; Fu and Farzaneh, 2010).
Ice accumulation along the blade is highly affected by variation of
centripetal and gravity forces along blade span sections which can cause
highly irregular ice shapes (Fu and Farzaneh, 2010; Bose, 1992b). The
rate of ice accretion increases with water droplet size (Homola et al.,
2010) and atmospheric temperature significantly influences ice shape.
Streamlined ice shapes are formed at lower temperatures and horn ice is
formed at higher temperatures (Homola et al., 2010). Glaze ice condi-
tions can produce higher power loss compared to rime ice conditions, due
to the greater ice thickness and shape of glaze ice (Reid et al., 2013). An
investigation of ice accretion on four different wind turbines sizes indi-
cated that rime ice conditions are less severe for large wind turbines
compared in terms of ice mass and thickness (Virk et al., 2010b; Homola
et al., 2010). The most severe icing problems are typically from glaze ice
conditions during freezing rain and drizzle (Bose, 1992a), which can
produce higher losses in lift coefficient than rime ice conditions (Hudecz
et al., 2013).
Limited data is available on airfoil profile's effect on ice accretion, in
terms of ice quantity and shape, as well as the effect on aerodynamic
performance. To further examine ice accretion impacts on wind turbines,
numerical studies are performed to determine the connection between
icing parameters and blade designing parameters. In this paper, ice ac-
cretion on wind turbines is investigated in terms of the following:
 Airfoil type of the NREL airfoil families of S801 to S835 (inclusive)
 Blade thickness of 50%–150% of airfoil blade for NREL S809
 The effect of ice accretion on aerodynamic performance (lift and
drag) for different icing conditions
Studying blade design parameters such as blade thickness is useful to
investigate ice accretion physics at different locations along the blade
span, as typical turbine designs often have varying airfoil curvature/
thickness from the blade's root to tip. Lastly, the effect of different ice
shapes, caused by differing icing conditions, on aerodynamic perfor-
mance provides important insights for wind turbine design and

40
G.M. Ibrahim et al. Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52

Fig. 1. Solution scheme for (a) single-shot approach and (b) multi-shot approach.

 
placement strategies. The results of this paper provide valuable new in- 2
σ ij ¼ δij pa þ μa δjk ∇k vi þ δik ∇k vj  δij ∇k vk ¼ δij pa þ τij (2)
sights and new numerical predictions to design wind turbines that can 3
operate in harsh northern locations.
where pa and μa represent air static pressure and dynamic viscosity of air,
2. Formulation of multiphase flow and phase change

Typically, the numerical simulation of ice accretion mainly consists of

the following modules: air flow solution, droplets collection efficiency
calculations, evaluation of boundary layer characteristics, and ice mass
evaluation by thermodynamic model computation (Reid et al., 2013).

2.1. Flow field calculations

The CFD simulations are steady incompressible flow. The air flow
solution is solved by FENSAP solver with the 3D Navier-Stokes and en-
ergy equations (ANSYS, 2015). The Navier-Stokes equation is
expressed as
⇀
∂ρa Va ⇀

⇀ ⇀  ⇀
þ ∇ : ρa Va Va ¼ ∇ :σ ij þ ρa :g (1)
∂t

where ρa is the air density and Va is the velocity vector. The stress tensor,
σ ij , is represented by

Fig. 2. One of NREL airfoil families, S809 profile.

41
G.M. Ibrahim et al.
Fig. 3. Computational domain discretizations for NREL S809, (a) O grid domain elevation view, (b) cell distribution around the blade profile, and (c) 3-D span wise view.
respectively. The conservation of energy equation is
   
∂ρa Ea ⇀
⇀  ⇀ ⇀ ⇀ ⇀
þ ∇ : ρa Va Ha ¼ ∇: Ka ∇ Ta þ vi τij þ ρa g: Va
∂t
where E and H are the total internal energy and enthalpy, respectively.
The airflow around a blade is typically turbulent at high Reynolds
numbers, thus one-equation Spalart Allmaras turbulence model (Hudecz
et al., 2013), with low free stream turbulence (using a very low Eddy/-
laminar viscosity ratio of 1  105) is selected (ANSYS, 2015; Spalart and
Allmaras, 1992). The Spalart-Allmaras model was selected as a
compromise between acceptable computational cost and the required
accuracy in turbulent flow simulations (Virk et al., 2010a). The model
Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52
adds a single additional variable for an undamped kinematic eddy vis-
cosity. It is designed for a low Reynolds number and can compute the
entire domain flow field to the airfoil surface. The model was originally
(3)
created for aerodynamic applications and is beneficial for its moderate
resolution requirements, stability and convergence ability. The SIMPLE
algorithm is used for the pressure velocity coupling. The heat fluxes at the
blade wall are computed with second order accuracy by solving the en-
ergy equation. Lift and drag coefficients can be calculated from
2Fl
Cl ¼ (4)
ρa;∞ Va;∞
2 A
∞
42
G.M. Ibrahim et al. Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52

2Fd fluid. The variables T∞ ; V∞ ; LWC and β represent the airflow and droplets
Cd ¼ (5)
ρa;∞ Va;∞
2 A
∞ reference parameters. The collection efficiency is calculated by
⇀ ⇀
where Fl and Fd are lift and drag forces, respectively, and A∞ is the αVd :n
β¼ (11)
reference surface area (based on C, chord length and blade span-wise ðLWCÞV∞
length). The torque coefficient, Cy, is determined by
⇀
where n is surface normal vector. The flow and droplet solutions are
Cy ¼ Cl sin ϕ  Cd cos ϕ (6) input to the ICE 3D solver to predict ice accretion (ANSYS, 2015).
In this paper, single- and multi-shot solutions methodologies are
where ϕ is the angle between the relative wind direction and plane of employed. Single shot is selected for the study of ice and the airfoil type
blade rotation. because of its lower computational expense compared to multi-shot
computations. A single-shot methodology is useful for capturing the
2.2. Droplet calculations differences of ice mass with airfoil type, but less accurate at predicting
the localized ice accretion quantity for specific icing conditions. A multi-
Water concentration, droplet velocity vectors, collection efficiency, shot approach provides more accurate predictions for the localized ice
impingement, shadow zone properties and impingement limits over the accumulation as it accounts for the changes in droplet collection distri-
entire domain are provided by the DROP 3D solver (ANSYS, 2015). The bution that is caused by accumulated ice changing thing the aero-
general Eulerian two-fluid model is used, which consists of Navier-Stokes dynamics of the airfoil. In this paper, a multi-shot solution methodology
equations augmented by the droplets’ continuity and mo- is used for investigating the aerodynamic effects of ice accumulation.
mentum equations, For single shot calculations, flow field and droplet solutions are
∂α ⇀
⇀ 
computed once and icing solution is obtained for total icing time as one
þ ∇ : αVd ¼ 0 (7) interval (Fig. 1a). Flow field solutions are used as an input to the droplet
∂t
solver to compute droplet collision efficiencies and impingement pro-

⇀  files. Then droplet solutions are transferred manually to ICE 3D solver to
∂ αVd
⇀  C R
⇀    predict ice mass/shape for one interval of an icing time. For an accurate
⇀ ⇀ D ed ⇀ ρ 1
þ ∇ αVd  Vd ¼ α Va  Vd þ α 1  a prediction of ice shape, multi-shot approach is recommended. As for
∂t 24K ρd Fr2
multi shot calculations (Fig. 1b), the transient changes in geometry,
(8)
surface heat fluxes and the droplet impingement profile due to ice for-
mation are taken into account. The total icing time is divided into short
where α and Vd are water volume fraction and droplet velocity, respec-
time intervals and an icing solution is obtained for each interval (i.e. for
tively. The first term in the momentum equation represents the drag on a
total icing time of 1800 s the time interval for 3 shots equals 600 s)
droplet with MVD. The second term represents the buoyancy and gravity
(ANSYS, 2015). Number of shots means the number of divisions of a total
forces. Collection efficiency is a dimensionless parameter that describes
icing time and each time interval is an iteration. Flow field, droplet and
the effective collision forces of droplets on a blade surface. Its peak value
is typically near the flow stagnation point. The droplet distribution is set
as mono-disperse (uniform droplet diameters), which has provided ac- Table 1
curate results in previous papers (Virk et al., 2010a). Additionally, a Test conditions for validation of numerical predictions.
comparison of mono-dispersed and Langmuir-D distribution (which Parameter Quantity
contains 7 different droplet sizes) (ANSYS, 2015), indicated that the Chord length 0.53 m
composite solution for the distribution is fairly close to solution of the Air speed 67.1 m/s
mono disperse model. To represent the icing event, the median volu- LWC 1.0 g/m3
metric diameter (MVD) of the droplets is chosen as a single quantity. MVD 20 μm
AoA 4

Temperature 7.78
C
2.3. Icing calculations Icing time 6 min

The ICE 3D solver uses mass conservation and energy conservation

equations to solve complex thermodynamics of ice formation (ANSYS,
2015). The equation expressing mass conservation is
 
∂hf ⇀ 
ρf þ ∇ : V f hf ¼ V∞ LWCβ  m_ evap  m_ ice (9)
∂t

where f and β represent water film and collection efficiency, respectively.

The first term represents the mass of water impingement, the second term
is mass of evaporated water and the third term is ice mass. The conser-
vation of energy equation is
2 3 2

2 3

e
6∂hf cf Tf ⇀  7 

Vd

ρf 4 þ ∇ : V f hf cf T~ f 5 ¼ 4cf T~ ∞  T~ f þ 5V∞ LWCβ

∂t 2

 
Levap m_ evap þ Lfusion  cs T~ m_ ice þ σε T∞ 4  Tf 4  ch T~ f  T~ ice;rec
þ Qantiicing
(10)
Fig. 4. Ice accretion on NACA 0 012 obtained numerically and experimentally
(ANSYS, 2015).
where ρf ; cf ; cs ; ε; σ; Levap and Lfusion represent the physical properties of

43
G.M. Ibrahim et al. Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52

Fig. 5. Cross-section of NREL airfoil profiles investigated in this paper.

Table 2 Table 3
Test conditions to examine curvature impact on ice mass. Tested ranges of airfoil curvature parameters.

Parameter Quantity Curvature parameter Examined range (m)

NREL airfoil families From S801 to S835 (S824 data is not available) TMax 0.11496–0.26868
Wind speed (m/s) 75 XOffset 0.07983–0.2454
AoA (
) 0 XMean 0.251755–0.5107445
Icing temperature (
C) 4 XL 0.168058–0.546335
MVD value (μm) 30 XU 0.24461–0.508787
LWC value (g/m3) 0.5
Blade thickness 100% (normal value corresponds to 1 m chord length)
Icing time (minutes) 30

Fig. 6. Airfoil curvature characterizations. Fig. 7. The effect of airfoil geometric parameters on airfoil icing for a NREL airfoil.

ice calculations are computed for the first interval of time to predict first
ice layer formation on blade. A new computational grid is generated for
the formed geometry due to ice accretion. The generated grid is then used
as an input to predict the second ice layer and the process is continued for
each shot (ANSYS, 2015). In this paper, other solution methodologies are
maintained for single- and multi-shot simulations (e.g., turbulence
model, conservation equations, among others). For both single and
multi-shot approaches, a convergence criteria of 1010 for fluid flow

44
G.M. Ibrahim et al.
calculations and 1010 for the energy equations is maintained. For all
multi-shot simulations, full convergence is achieved for each shot.
2.4. Mesh discretizations
The NREL airfoil families are selected to investigate icing as they are
widely used for HAWTs blade industry (Tangier and Somers, 1995). For
all airfoils, a chord length of 1 m is maintained along a blade span of
0.25 m to represent a small portion of a rotor blade near the tip. A span of
0.25 was shown in a previous study to provide useful results (Virk et al.,
2010a). A fine 3D structured grid is created for each airfoil with identical
discretization schemes. As for a NREL S809 profile (Fig. 2), a fine 3D
structured grid with 35,280 cells is created with a minimum orthogonal
quality of 0.835 (where values close to 0 correspond to low quality and 1
represents high quality) and maximum orthogonal skew of 0.194 (where
values close to 1 correspond to low quality and 0 represents high quality).
The yþ value is 0.2. The grid's first boundary layer thickness over the
blade is 1  106 m with a growth rate of 1.1 for 98 layers to accurately
capture ice shape and mass on blade surface. The main advantages of
using a structured grid are higher resolution and better accuracy. The
structured domain uses hexahedra cells which are characterized by reg-
ular connectivity and efficient spacing. As shown in Fig. 3a, a circular
domain (O grid type) with a diameter of 20 m is created. The blade span
Fig. 8. Ice mass variations with TMax/XMean.
Fig. 9. Droplets impingement profile (droplet collection) for 3 different blade sizes.
45
Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52
is discretized into one cell of 0.25 m and along the chord there are 360
discretizations, in a mapped geometric progression scheme. Ice accretion
is mostly formed at the leading edge due to droplet collection efficiency.
Smaller elements are created at the leading edge with equal spacing of
0.001 m to accurately evaluate both ice mass and shape. The cell size
gradually changes along the airfoil surface, as illustrated in Fig. 3b and c.
The grid independence study investigated different mesh resolutions,
approximately 25,000, 35,000 and 45,000 elements, at a specific icing
condition resulting in minimal changes in ice mass (approximately ± 4 g).
Lift and drag values are obtained after convergence with minimal dif-
ferences of ± 0.0001 in their quantities. Convergence is achieved for all
variables (i.e., momentum, mass, energy and turbulence model equations
and droplet solution residuals).
3. Results validation
A study (Shin and Bond, 1992) on icing of NACA 0 012 airfoils using
LEWICE code and wind tunnel tests is used to validate this work in terms
of ice shape. A computational grid is created around NACA 0 012 with
the same process and configurations used for NREL S809. As presented in
Table 1, one condition from the experiment (Shin and Bond, 1992) is
investigated using FENSAP ICE solver, to obtain an ice shape comparison.
As illustrated in Fig. 4, the predicted ice shapes agree well with
Fig. 10. Total droplet collection with NREL blades of various thicknesses.
Fig. 11. Numerically predicted effect of blade thickness and droplet collection on accreted
ice mass on a HAWT blade.
G.M. Ibrahim et al.
previous numerical and experimental data. The total icing time is divided
in 5 intervals (72 s) to evaluate an accurate ice shape for comparison. For
each interval of time, one ice shot/layer is added. Then the grid
displacement is obtained and transferred to start the calculations for the
second shot taking into account transient changes happening in flow
field, droplet impingement profile. For 5 ice shots, the ice horn thickness
(approximately x/c ¼ 0.025) at the leading edge is nearly equal to the
one predicted by experimental investigation (Shin and Bond, 1992).
Minimal changes (especially after the third shot) in ice shapes are pre-
dicted. The more shots the accurate ice shapes and high computational
time to obtain simulated ice shapes. Therefore, based on nearly equal ice
shapes when comparing FENSAP ICE data with experimental data, ice
mass is nearly identical and 3 shots is suitable for this paper.
For aerodynamic efficiency prediction due to ice accumulation, the
experimental results analysed by Broeren and Bragg (Spalart and All-
maras, 1992; Tangier and Somers, 1995) at a Reynolds number of
6.5  106 m showed that the reduction in the aerodynamic efficiency
can range from 15% to 80% due to ice accretion. The current predicted
Fig. 12. Ice loads on NREL airfoil families for HAWTs.
Table 4
Test conditions to examine blade thickness effect at various icing conditions.
46
Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52
results in this paper agree well with the experimental results by Broeren
and Bragg Broeren et al. (2004); Hor
ak et al.(2008), with a reduction in
lift ranging from 10% to 65%.
4. Results and discussion
In this section, numerical predictions of ice accretion on HAWTs with
NREL airfoils and the effect on aerodynamic performance are presented.
The analysis is in terms of (i) the quantity of ice accretion for 35 different
NREL airfoils (S801 to S835, inclusive), (ii) the quantity of ice accretion
for a NREL airfoil S809 for different blade thicknesses, and (iii) the effect
of ice accretion on a NREL airfoil S809 at different angles of attack.
Significant parameters that are important for this study include wind
speed/relative wind velocity, air temperature, LWC, MVD, blade curva-
ture parameters, angle of attack (AOA), blade surface roughness and
icing time, among others. In the numerical predictions, the following
operating conditions are investigated.
 Relative wind velocity - a relative wind speed at a span wise location
80% from the rotor blade root is represented, as 80% from the rotor
blade root is typical location of high ice accretion (Virk et al., 2010a).
Therefore, assuming a TSR of 6 and a wind speed of 15.3 m/s, a
relative wind speed of 75 m/s is calculated and maintained for all
simulations in this paper. A relative wind speed of 75 m/s can also be
calculated from a TSR of 7 and a wind speed of 13.2 m/s.
 Air temperature and blade surface temperature are assumed uniform
(i.e., air temperature ¼ blade surface temperature). Two tempera-
tures are selected, 4
C and 8
C, to represent temperatures that
can cause glaze and rime icing conditions, respectively.
 Current wind turbine towers can reach up to 135 m, and blade tip
height can reach 200 m above ground. Atmospheric icing conditions
mainly depend on the surrounding environment with lowest cloud
extent at 480 m above sea level for cumuliform clouds and more than
1 km for stratiform clouds. As LWC changes with altitude, the
assumption of no significant change in values of LWC and MVD is
maintained for atmospheric icing of wind turbines (Fikke et al.,
2006). Since Meteorological data at lower altitudes (wind towers) are
not available, different icing conditions with higher and lower LWC
G.M. Ibrahim et al. Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52

Fig. 13. Relationship between LWC and MVD for the examined conditions of atmospheric
icing on wind turbines.

concentrations can be numerically predicted using cloud

characteristics.
 The LWC and MVD values investigated in blade section effect are
selected from measured values of LWC variation with MVD at
different temperatures, which is in the DROP 3D solver. The selection
is based on different cloud conditions (stratiform cloud for low con-
centrations and cumuliform for higher ones) (Jeck, 2002), to provide
conditions of LWC that exist at specific MVD. For lower and higher
LWC concentrations, stratiform and cumuliform clouds conditions at
lowest altitude are applied to investigate different icing events. This
method captures the impact of different icing conditions, in terms of
high and low LWC, at specific MVD ranges, for a specified
temperature.
 Icing time - an icing time of 30 min is simulated, which is a critical
time for determining whether to stop turbine operation. Fig. 14. Predicted ice accretion on different NREL S809 airfoil thicknesses for (a) low
 Blade roughness - the sand-grain roughness model is used and LWC conditions and (b) high LWC conditions.
maintained at 0.0005 m, which has been shown to be an accurate
representation for typical icing predictions based on experimental
validation (ANSYS, 2015). Surface roughness is a significant factor in Table 5
predicting/calculating heat fluxes and shear stresses for icing simu- Test conditions to examine blade aerodynamics during icing conditions.

lations, although not the focus of this paper. Parameter Quantity

 Blade curvature - parameters are investigated in terms of the airfoil Airfoil designated NREL S809
maximum thickness and its location. name
 Airfoil thickness (Section II of the Results and Discussion) - the NREL AOA (
) 0
S809 airfoil is selected to study the effect of blade thickness variation Wind speed (m/s) 75
Icing temperature 4, 8
on ice mass during different icing events characterized by high and (
C)
low LWC conditions. MVD value (μm) 30
 To examine the airfoil design or curvature impact on ice mass all LWC value (g/m3) 0.5, 1
NREL families are used. As ice accumulates on blade surface airfoil Blade thickness 0.21 m (Normal thickness corresponds to a chord length of
1 m)
curvature changes noticeably causing a significant impact of aero-
Icing time (minutes) 30
dynamics, therefore various ice shapes at different icing conditions
are used to study this impact.
YL, XOffset, TMax, and XMean. The variable XOffset is the offset distance be-
5. The effect of airfoil shape on ice accumulation quantity for a tween the points (XU,YU) and (XL,YL), TMax is blade maximum thickness
HAWT with an NREL airfoil and XMean is the average offset distance from the origin (TMax location on
X axis). Blade curvature/leading edge radius is highly affected by the
In this section, ice accumulation is investigated for the NREL airfoil location of these two points (XU, YU) and (XL, YL) for both upper and
families. As shown in Fig. 5, all NREL airfoil families are analysed to lower sides, respectively. Table 3, shows the examined ranges of curva-
study the relationship between their curvature's significant parameters ture parameters for NREL airfoil families analyzed in this paper.
and ice accretion. As presented in Table 2, one icing condition is main- The combined effects of the airfoil's geometry on the flow stream and
tained to investigate the effect of blade design on ice accumulation. ice accretion are difficult to distinguish. The geometric variables that can
As illustrated in Fig. 6, airfoil curvature can be defined by XU, YU, XL, be used to identify an airfoil's profile were systematically analyzed based

47
G.M. Ibrahim et al.
Fig. 15. Numerically predicted ice shapes for different icing conditions at (a) LWC ¼ 0.5 g/m3, T ¼ 4
C, (b) LWC ¼ 1 g/m3, T ¼ 4
C, (c) LWC ¼ 0.5 g/m3, T ¼ 8
C and (d)
LWC ¼ 1 g/m3, T ¼ 8
C.
on the numerical predictions of ice accumulation. As illustrated in Fig. 7,
the geometric variable TMax and YMean have the most significant effect on
ice accumulation. The vertical axis is the ice accumulation and the hor-
izontal axis represents a normalized geometric variable, whereby the
values are divided by their highest quantity. The trend line's slope for
TMax, YMean, and YOffset are 0.56, 0.44, and 4.3, respectively, with a co-
efficient of determination (r2) of 0.79, 0.82, and 0.74 (representing
medium to high correlation), respectively. The geometric variables XMean
and XOffset show only a minor correlation with ice accumulation with
trend line slope's of 0.36 and 0.07, respectively and coefficient of de-
terminations of 0.17 and 0.03 respectively (representing low
correlation).
As illustrated in Fig. 8, the location of maximum thickness has a
significant impact on ice load and can be represented by plotting ice mass
with TMax/XMean. Higher values of TMax/XMean ratio suggests that the
location of maximum thickness is near the leading edge causing greater
ice accumulation. Locating XOffset (distance between the two points
(XU,YU) and (XL,YL)) near the blade's leading edge can lead to a large
stagnation area and increased ice accumulation. However, locating XOffset
near the blade's trailing edge can reduce the stagnation area and decrease
the quantity of accumulated ice.
In Fig. 9, droplets collection efficiencies are illustrated for three
different airfoils to show how the impingement area is affected due to
blade thickness. For thinner airfoils (e.g., S806), the collection efficiency
profile shows the lowest y/c (0.02 to 0.04), while thicker airfoils (e.g.,
S821) have the largest y/c (0.1 to 0.06), which illustrates a contributing
factor to increased ice accumulation for thicker airfoils. The peak value of
collection efficiency can be found near the point of flow stagnation. The
48
Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52
curve profile is not identical for the pressure side and suction side of a
blade section because of asymmetry of the airfoils. Higher collection
efficiency peak magnitude can occur for thinner blade sections, thus ice
accretion is more affected by the total collection efficiency rather than its
peak value.
As illustrated in Fig. 10, total droplet collection has a linear trend
(R2 ¼ 0.7405) with NREL blade thickness. The total droplet collection is
defined as the integral of collection efficiency (i.e., total area under the
collection efficiency curve, Fig. 9), which represents the quantity of
droplets impacting the wind turbine blade. The integral values are ob-
tained for all NREL airfoils and suggest that blade thickness has a sig-
nificant effect on the quantity of droplets colliding with the blade surface.
Furthermore, as illustrated in Fig. 11, total droplet collection is nearly
perfectly correlated (R2 ¼ 0.9997) with accreted ice mass, thus accreted
ice mass on a wind turbine blade can be predicted based on the airfoil's
thickness, with moderate agreement (R2 ¼ 0.74).
As illustrated in Fig. 12, ice load significantly increases with airfoil
thickness as thicker airfoils (e.g., S804, S808, S815, S821) tend to have
greater ice accumulation compared to thinner airfoils (e.g., S802, S806,
S817, S832). The differences in ice quantities between thicker
(S821 ¼ 0.8) and thinner (S806 ¼ 0.43) airfoils is significant. Thick
blades tend to have more droplet collection area at the leading edge, thus
the droplet impingement area is large and can cause greater ice accu-
mulation. Differences between the upper and lower points on the X-axis
(XOffset) or maximum thickness location, can lead to various stagnation
area configurations, which is the likely cause of airfoils with lower
thickness having slightly equal or large ice quantity (i.e. comparing S802
with S803, S808 with S811 and S821 with S823). For example, S823
G.M. Ibrahim et al. Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52

Fig. 16. Computational grids around airofils with accreted ice for different icing conditions: (a) LWC ¼ 0.5 g/m3, T ¼ 4
C, (b) LWC ¼ 1 g/m3, T ¼ 4
C, (c) LWC ¼ 0.5 g/m3, T ¼ 8
C
and (d) LWC ¼ 1 g/m3, T ¼ 8
C.

(0.786) is having nearly equal ice mass with S821 (0.785), however S821
have comparatively higher thickness, this can highlight the significance
of other curvature important parameters like TMax/XMean ratio or
maximum thickness location.
6. The effect of airfoil thickness on ice accumulation quantity for
a HAWT with an NREL S809 airfoil
This section presents predicted ice accumulation on five different
thicknesses of the same airfoil type, NREL S809, to investigate the

49
G.M. Ibrahim et al. Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52

significance of blade thickness on ice load. For lower MVD, below 20 μm,
differences in ice mass do not change to the same extent as larger ones,
highlighting the significance of MVD during ice accumulation. For larger
droplet sizes, ice accretion variation with blade thickness is significant.
Icing environments with small droplets at high and low LWC concen-
trations has a minor impact on ice amount for different blade thicknesses
compared to large droplets. LWC content have a significant impact on ice
amount as it can be shown when comparing Fig. 14a with 14c, that ice
mass values is increased at fixed droplet size comparison.

7. The effect of ice accretion on aerodynamic performance for a

HAWT with an NREL S809 airfoil

Aerodynamic characteristics significantly depend on blade curvature.

During an ice accretion process, the blade profile changes with time,
changing both lift and drag characteristic. As shown in Table 5, different
icing conditions are examined to numerically obtain various ice shapes.
This comparison between clean and iced profiles is used to predict the
changes in aerodynamic performance.
As ice accretes on the blade surface, aerodynamics change and
considerably affect flow field, heat fluxes and droplet impingement dy-
namics. Multi-shot approach in FENSAP ICE solver accounts for the
transient changes occurring during an ice formation process. Thus, multi-
shot approach is used to accurately predict ice shapes for this analysis.
As illustrated in Fig. 15, ice shapes at 4
C and 8
C temperatures
with two LWC conditions are numerically predicted using a multi-shot
approach. Generally, icing shape and location are controlled by flow
speed and angle of attack. Ice thickness at the stagnation point of the flow
is highly influenced by the atmospheric temperature, as ice thickness at
this point is increased when comparing ice shapes at 4
C and 8
C,
under fixed LWC and MVD conditions.
At 4
C, LWC significantly affects the quantity of ice accumulation,
as well as its propagation along the blade section. At 8
C, a horn ice
shape is formed and its height is affected by LWC, as a significantly more
pronounced ice horn accumulates at a LWC of 1 g/m3 compared to 0.5 g/
m3. As illustrated in Fig. 16, a computational grid around an iced NREL
S809 blade is created by the grid displacement tool existed in the solver
for each ice shape to numerically predict the lift and drag coefficients.
The effect of the four ice shapes (Fig. 15) on aerodynamic perfor-
mance, is investigated by analysing the lift and drag at several angles of
Fig. 17. Aerodynamic performance for predicted ice shapes, in terms of (a) lift coefficient
and (b) drag coefficient.
attack (i.e., the ice shapes are formed at an AoA of 0
, and then the AoA is
changed to study the effect of ice on lift and drag). As illustrated in
Fig. 17, ice accretion can produce degradation of the lift coefficient, as
relationship between blade thickness and icing conditions in terms of
well as increase drag. Every ice distribution can cause different flow
LWC and MVD at specific temperature. As presented in Table 4, blade
characterises around the blade, including location of flow stagnation
thickness is varied from 50% to 150% (where 100% value represents the
point (affected by angle of attack) and flow separation behaviour
original equivalent blade thickness for a chord length of 1 m), thus 50%
(affected by height/thickness of the ice horn). A significant reduction in
thickness represents a thin blade and 150% thickness represents a
lift, ranging from 10% to 65%, is predicted. Furthermore, the increase in
thick blade.
drag and decrease in lift is substantial for large horn ice. As wind turbine
Flow field solutions are computed by FENSAP solver, and then droplet
blade lift decreases the overall torque will decrease leading to a reduction
calculations for each icing conditions (LWC and MVD) are obtained using
in output power.
DROP 3D solver. Finally, icing calculations are predicted for each con-
In Fig. 18, flow fields (in terms of velocity magnitude) are presented
dition using a single shot approach. Icing conditions with low and high
for the clean blade and four iced blades, at an AoA of 8
. The larger ice
LWC conditions are investigated. Typically, smaller droplet sizes exist at
shape formation causes more flow separation (e.g., comparing ice shape
higher LWC conditions while large droplets are at lower LWC conditions
(d) with a clean blade). As flow behaviour around the blade changes due
(Jeck, 2002).
to the formed ice, the overall lift force from the blade section is signifi-
As illustrated in Fig. 13, eight different LWC/MVD ratios are selected
cantly reduced. The flow fields show the drastic changes the accumulated
to represent various atmospheric icing conditions. The eight different
ice has on the airfoil aerodynamics, which can significantly impair
LWC quantities, at a temperature of 4
C, from 0.17 g/m3 to 3.7 g/m3
HAWT performance.
(selected as discussed above from appendix C) are investigated, which
To determine the effect of reduced lift and increased drag due to ice
represent icing events with low and high LWC conditions. Eight different
accumulation on the performance of a HAWT, the torque coefficient is
MVD values, from 15 μm to 50 μm, are investigated at a temperature of
determined for various AoA for the clean and the four iced blades. To
4
C, which is useful to study the impact of droplet size during ice
determine the angle of relative wind, a 40 m blade radius is assumed,
formation on blade profile.
therefore, at 80% blade radius (32 m), the torque coefficient is deter-
As illustrated in Fig. 14, for larger blade thicknesses higher quantities
mined. A TSR of 6 and a wind speed of 15.3 m/s is also assumed. As
of ice are formed for differing icing conditions, which highlights the
illustrated in Fig. 19, the different ice shapes have severe impacts on

50
G.M. Ibrahim et al. Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52

Fig. 18. Predicted velocity magnitude contours for clean and iced NREL S809 airfoils on a HAWT.

HAWT performance. The optimal relative angle of attack for best aero-
dynamic performance is at a different location, depending on the
ice shape.
8. Conclusions
In this paper, ice accretion on a wind turbine blade was investigated
using FENSAP ICE solver. The predicted results identified that droplet
size has a significant effect on ice accumulation for fixed temperature and
LWC conditions. The results of this paper provide the following
key insights.

 From predictions on ice loads on NREL airfoil families, it was shown

that blade thickness and its location from the leading edge has a
greater impact on ice quantity than any other blade curvature
parameter.
 Thicker blades have a collection efficiency characterized by large
coverage area (y/c) at the leading edge when compared to thinner
ones, thus ice loads increase significantly with blade thickness.
 Accreted quantity of ice is dependent on the blade thickness due to
the increased droplet collection of thicker blades
 Ice mainly formed/found at the blade's leading edge and its location is
controlled by angle of attack.
 Atmospheric icing temperature affects ice thickness at the location of
flow stagnation, as ice thickness at this location is increased signifi-
cantly when comparing ice shapes at 4
C and 8
C, under same
icing conditions.
 Propagation of the accreted ice along blade section is expected for an
icing temperature of 4
C, with higher LWC conditions.
Fig. 19. The effect of ice accumulation on aerodynamic performance of a HAWT.

51
G.M. Ibrahim et al.
 At a temperature of 8
C or below, horn ice shapes are expected, its
thickness and height are controlled by LWC concentration. Aero-
dynamic behaviour prediction mainly depends on accreted ice shape
on the blade's surface and ice accretion significantly increased drag
coefficient.
 The loss in lift coefficient due to ice accretion was between 10% and
65%, depending on accreted ice shape.
Acknowledgments
The authors of this paper gratefully acknowledge the financial sup-
port of the Research and Development Corporation of Newfoundland
and Labrador.
References
ANSYS, 2015. FENSAP-ICE User Manual R1.0, Canada.
Battisti, L., 2015. Wind Turbines in Cold Climates: Icing Impacts and Mitigation Systems.
Springer.
Borg, M., Shires, A., Collu, M., 2014. Offshore floating vertical axis wind turbines,
dynamics modelling state of the art. Part I: aerodynamics. Renew. Sustain. Energy
Rev. 39, 1214–1225.
Bose, N., 1992. Icing on a small horizontal-axis wind turbine - Part 1: glaze ice profiles.
J. Wind Eng. Industrial Aerodynamics 45 (1), 75–85.
Bose, N., 1992. Icing on a small horizontal-axis wind turbine - Part 2: three dimensional
ice and wet snow formations. J. Wind Eng. Industrial Aerodynamics 45 (1), 87–96.
Broeren, A.P., Bragg, M.B., Addy, H.E., 2004. Effect of intercycle ice accretions on airfoil
performance. J. Aircr. 41 (1), 165–174.
Cattin, R., 2013. Icing of Wind Turbines, Vindforsk Projects, a Survey of the Development
and Research Needs, Elforsk Report 13.
Cober, S.G., Isaac, G.A., Strapp, J.W., 2001. Characterizations of aircraft icing
environments that include supercooled large drops. J. Appl. Meteorology 40 (11),
1984–2002.
Fikke, S.M., Ronsten, G., Heimo, A., Kunz, S., Ostrozlik, M., Persson, P.E., Laakso, T.,
2006. Atmospheric Icing on Structures: Measurements and Data Collection on Icing:
State of the Art, Meteo Schweiz.
Fu, P., Farzaneh, M., 2010. A CFD approach for modeling the rime ice accretion process
on a horizontal axis wind turbine. J. Wind Eng. Industrial Aerodynamics 98 (4),
181–188.
Habashi, W.G., Morency, F., Beaugendre, H., December 2001. FENSAP-ICE: a
comprehensive 3D simulation tool for in-flight icing. In: 7th International Congress of
Fluid Dynamics and Propulsion, Sharm-El-Sheikh, Egypt.
H
e, lo-iacute, Beaugendre, S., Fran-atilde, Morency, O., Habashi, W.G., 2003. FENSAP-
ICE's three-dimensional in-flight ice accretion module. ICE3D. J. Aircr. 40 (2),
239–247.
52
Journal of Wind Engineering & Industrial Aerodynamics 173 (2018) 39–52
Hochart, C., Fortin, G., Perron, J., Ilinca, A., 2008. Wind turbine performance under icing
conditions. Wind Energy 11 (4), 319–333.
Homola, M.C., Virk, M.S., Wallenius, T., Nicklasson, P.J., Sundsbø, P.A., 2010. Effect of
atmospheric temperature and droplet size variation on ice accretion of wind turbine
blades. J. Wind Eng. Industrial Aerodynamics 98 (12), 724–729.
Hor
ak, V., Rozehnal, D., Ch
ara, Z., Hyll, A., 2008. CFD and Experimental Study of
Aerodynamic Degradation of Iced Airfoils.
Hudecz, A., Koss, H., Hansen, M.O.L., 2013. Ice accretion on wind turbine blades. In: 15th
International Workshop on atmospheric icing of structures (IWAIS XV).
Jeck, R.K., 2002. Icing Design Envelopes (14 CFR Parts 25 and 29, Appendix C) Converted
to a Distance Based Format (No. DOT/FAA/AR-00/30). Federal Aviation
Administration Technical Center, Atlantic City, NJ.
Kraj, A.G., Bibeau, E.L., 2010. Phases of icing on wind turbine blades characterized by ice
accumulation. Renew. Energy 35 (5), 966–972.
Makkonen, L., 2000. Models for the growth of rime, glaze, icicles and wet snow on
structures, Philosophical Transactions of the Royal Society of London A:
Mathematical. Phys. Eng. Sci. 358 (1776), 2913–2939.
Makkonen, L., Laakso, T., Marjaniemi, M., Finstad, K., 2001. Modelling and prevention of
ice accretion on wind turbines. Wind Eng. 25 (1), 3–21.
Marjaniemi, M., Makkonen, L., Laakso, T., 2000. TURBICE-the wind turbine blade icing
model. In: In Proceedings of the BOREAS V conference.
Masters, C.O., 1985. A New Characterization of Super Cooled Clouds below 10,000 Feet
AGL.
Paulsen, U.S., Madsen, H.A., Kragh, K.A., Nielsen, P.H., Baran, I., Hattel, J.,
Berthelsen, P.A., 2014. DeepWind-from idea to 5 MW concept. Energy Procedia 53,
23–33.
Reid, T., Baruzzi, G., Ozcer, I., Switchenko, D., Habashi, W., January, 2013. FENSAP-ICE
simulation of icing on wind turbine blades, part 1: performance degradation. In: 51st
AIAA Aerospace Sciences Meeting Including the New Horizons Forum and Aerospace
Exposition, Grapevine, Texas.
Rezaeiha, A., Pereira, R., Kotsonis, M., 2017. Fluctuations of angle of attack and lift
coefficient and the resultant fatigue loads for a large horizontal axis wind turbine.
Renew. Energy 114 (B), 904–916, 2017.
Ruff, G.A., Berkowitz, B.M., 1990. User's Manual for the NASA Lewis Ice Accretion
Prediction Code (LEWICE).
Shin, J., Bond, T.H., 1992. Experimental and Computational Ice Shapes and Resulting
Drag Increase for a NACA 0012 Airfoil.
Spalart, P., Allmaras, S., January, 1992. A One-equation Turbulence Model for
Aerodynamic Flows. 30th Aerospace Sciences Meeting and Exhibit.
Tangier, J.L., Somers, D.M., 1995. NREL Airfoil Families for HAWTs National Renewable
Energy Laboratory, pp. 117–123.
Virk, M.S., Homola, M.C., Nicklasson, P.J., 2010. Relation between angle of attack and
atmospheric ice accretion on large wind turbine's blade. Wind Eng. 34 (6), 607–613.
Virk, M.S., Homola, M.C., Nicklasson, P.J., 2010. Effect of rime ice accretion on
aerodynamic characteristics of wind turbine blade profiles. Wind Eng. 34 (2),
207–218.
Virk, M.S., Homola, M.C., Nicklasson, P.J., 2012. Atmospheric icing on large wind turbine
blades. Int. J. Energy Environ. 3 (1), 1–8.