See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/223448025

An investigation of ship airwakes using Detached-Eddy Simulation

Article in Computers & Fluids · April 2010

DOI: 10.1016/j.compfluid.2009.11.002

CITATIONS READS

176 2,850

2 authors:

James Forrest Ieuan Owen

Prism Defence University of Liverpool
22 PUBLICATIONS 607 CITATIONS 137 PUBLICATIONS 2,601 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Ieuan Owen on 30 October 2017.

The user has requested enhancement of the downloaded file.

 Computers & Fluids 39 (2010) 656–673

Contents lists available at ScienceDirect

Computers & Fluids

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m p fl u i d

An investigation of ship airwakes using Detached-Eddy Simulation

James S. Forrest *, Ieuan Owen
Department of Engineering, University of Liverpool, Brownlow Hill, Liverpool L69 3GH, UK

a r t i c l e i n f o a b s t r a c t

Article history: Computational Fluid Dynamics simulations of ship airwakes have been performed using Detached-Eddy
Received 9 March 2009 Simulation (DES) on unstructured grids. A generic simple frigate shape (SFS2) and a Royal Navy Type 23
Received in revised form 5 June 2009 Frigate (T23) have been studied at several wind-over-deck (WOD) conditions. A comprehensive valida-
Accepted 5 November 2009
tion exercise has been performed, comparing CFD results of the airwake calculated for the SFS2 with high
Available online 12 November 2009
quality wind tunnel data provided by the National Research Council of Canada. Comparisons of mean
quantities and velocity spectra show good agreement, indicating that DES is able to resolve the large-
Keywords:
scale turbulent structures which can adversely impact helicopter–ship operations. An analysis of the air-
Computational Fluid Dynamics
Airwake
wake flow topology at headwind and Green 45° conditions highlights the dominant flow features over the
Detached-Eddy Simulation flight deck and it is shown that significant differences exist between the two WOD angles. T23 airwake
Frigate data has been compared to full-scale experimental results obtained at sea. It is shown that the inclusion
Bluff-body flow of an atmospheric boundary layer velocity profile in the CFD computations improves the agreement with
Validation full-scale data. Qualitative comparison between the simple frigate shape and T23 airwakes shows that
large-scale flow patterns are similar; but subtle differences exist, particularly at more oblique WOD
angles.
Ó 2009 Elsevier Ltd. All rights reserved.

1. Introduction gard is the modelling of ship airwakes. It has been suggested that
high fidelity piloted flight simulation could also be used to supple-
The launch and recovery of helicopters to naval ships poses sig- ment SHOLs obtained at sea [2], providing a much safer alternative
nificant challenges for the pilot. Landing decks are typically small to sea trials and giving the added benefit of allowing investigative
and are often moving due to the pitch, roll, and heave motions of flight tests while new ships are still at the design stage. One of the
the ship. Air passing over the ship’s superstructure due to a combi- main barriers to these advances is a lack of confidence in the accu-
nation of its forward speed and the prevailing wind causes the for- racy of airwakes currently deployed in flight simulators. The accu-
mation of a region of disturbed flow over the flight deck known as rate representation of time-varying disturbances resulting from
the ship airwake. Due to bluff-body flow separation and the com- airwake turbulence is a key factor in replicating workload levels
plex interaction of unstable separating shear layers and vortices, experienced at sea [3], however this is proving to be a particularly
the airwake contains time-varying turbulent structures which challenging aspect of DI modelling and simulation.
can have a significant impact on aircraft handling qualities. At cer- One approach to modelling ship airwakes is through the use of
tain wind-over-deck (WOD) conditions the airwake can cause suf- Computational Fluid Dynamics (CFD). In this method CFD is used to
ficient pilot workload to reduce the probability of a safe landing to solve for the flow over the ship geometry, with the resulting veloc-
unacceptable levels. To reduce the risks associated with line pilots ity field data exported to a flight simulation environment as look-
operating helicopters from ships, operators must develop ship– up tables. Early work in this field solved the steady-state equations
helicopter operating limits (SHOLs) through hazardous and time- of fluid flow [4,5], providing airwakes which consisted of mean
consuming First of Class Flight Trials (FOCFTs) [1]. velocity gradients. In [4] turbulent fluctuations were applied with-
Modelling and simulation of the ship-helicopter dynamic inter- in the simulator based on local flow gradients, however this can
face (DI) has been an active area of research in recent years. The only be classed as a first-order approach as the physical processes
use of flight simulation to aid pilot training for deck landings is responsible for turbulence generation are not represented. It has
now commonplace, but its effectiveness is highly dependent on been suggested that time-accurate CFD simulations which can cap-
simulator fidelity. One of the key areas for improvement in this re- ture the unsteady effects of ship airwake turbulence may provide
the required fidelity [6,7]. Such computations, which typically re-
quire thousands of hours of CPU time, have recently become feasi-
* Corresponding author. Tel.: +44 (0) 151 794 4692.
ble due to advances in computing power and, in particular, the
E-mail addresses: [email protected] (J.S. Forrest), i.owen@liverpoo-
l.ac.uk (I. Owen). increased availability of parallel computing. Numerous studies of

0045-7930/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compfluid.2009.11.002
 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673 657

Nomenclature

b ship beam (m) zref reference height (m)

d domain depth (m) C DES DES length scale constant
fb SST k  x eddy-viscosity constant I turbulence intensity, normalised by V 1 (%)
h hangar height (m) Lt turbulent length scale (m)
l deck length (m) Re Reynolds number
ls ship length (m) St Strouhal number
r domain radius (m) U local velocity magnitude (m/s)
Dt non-dimensionalised time-step U ref velocity magnitude at reference height (m/s)
u longitudinal velocity (m/s) V1 freestream velocity (m/s)
v lateral velocity (m/s) a surface roughness constant
w vertical velocity (m/s) D local grid spacing (m)
x longitudinal distance (m) D0 grid spacing in focus region (m)
y lateral distance (m) k2 second eigenvalue of S2 þ X2
yþ non-dimensionalised wall distance
z vertical distance (m)

time-accurate ship airwakes using the CFD code COBALT have been
performed by Polsky at NAVAIR [8,9] in support of the US Depart-
ment of Defense Joint Shipboard Helicopter Integration Process
(JSHIP) project [10]. Using the JSHIP Dynamic Interface Modeling
and Simulation System (DIMSS), a comprehensive series of simu-
lated flight tests were performed by four pilots flying the UH-
60A Black Hawk to an amphibious assault class (LHA) ship [11].
It was shown that the inclusion of the unsteady component of
the airwake significantly increased pilot workload during ship-
board manoeuvres. Lee et al. also generated unsteady CFD airwake
data for the LHA using the parallel flow solver PUMA/PUMA2 [12–
14]. In these cases the data was used for a shipboard approach and
landing task driven by an optimal control model of a human pilot.
Fig. 1. SFS2 geometry; extent of original SFS geometry shown by darker shading.
During these studies, it was also found that the time-varying air-
wake effects had an impact on the control activity predicted by
the pilot model. the flow field around a 1:100 model of the SFS2. Mean velocity data
Although the use of time-accurate airwake data has been shown and turbulence statistics were obtained at a number of locations
to be an important factor in improving the fidelity of DI simula- over the flight deck and the forward superstructure; long acquisi-
tions, it is equally important that the spatial and temporal charac- tion periods also allowed accurate velocity spectra to be derived.
teristics of turbulence introduced to the simulator are represented Several authors have published the results of computational
accurately. CFD methods are routinely validated using full-scale studies on the SFS/SFS2 geometries. Reddy et al. [21] used the com-
experimental data obtained in situ, or model-scale data from the mercial CFD code, FLUENT, with the k  e turbulence model to
wind tunnel. However, many of the published airwake studies con- solve the steady-state flow around the SFS on structured grids. Sig-
tain inadequate validation. Where attempts are made to compare nificant variations in flow topology were seen for different grid
CFD results with experimental data, the comparison is often qual- densities, although grid independence was approached as the cell
itative (e.g. [15]), turbulent characteristics are frequently neglected count rose above 1 million. Reasonable qualitative agreement with
(e.g. [16]) and in many cases the analysis is limited to a single WOD oil flow visualizations was shown, however the locations of reat-
condition (e.g. [4,17]). Sometimes the lack of detail is due to the tachment points and vortices differed from the experiments. Liu
sensitive nature of military naval hardware, which means that et al. [22] also analyzed the SFS, obtaining inviscid, unsteady re-
there has been a relative lack of high quality experimental data sults from mean flow field computations using non-linear distur-
made available for ship airwake validation studies. bance equations (NLDE). Calculations were performed using
In an effort to develop a ship airwake validation database, and parallel processing on structured grids and identified regions of
to facilitate the dissemination of best-practices amongst the DI high turbulence and vortex shedding over the flight deck. As part
simulation community, a collaborative ship airwake modelling of an analysis of rotor loads in ship airwakes, Wakefield et al.
activity was set up under the auspices of The Technical Co-opera- [23] computed flow over the SFS using a steady-state Navier–
tion Programme (TTCP) [18]. The simple frigate shape (SFS), a Stokes solver. Many of the large-scale flow features seen in the
highly simplified ship geometry, was created to provide an easily wind tunnel were replicated by the CFD, although at certain condi-
repeatable benchmark case for validating CFD codes. Fig. 1 shows tions the computational results predicted flow separation over re-
an updated version of this geometry, the SFS2, with an elongated gions of the flight deck which was not observed experimentally.
superstructure and a pointed bow. The National Research Council Roper et al. [5] performed validation studies of both the SFS and
of Canada (NRC) has performed a series of wind tunnel tests on SFS2 before the resultant ship airwake data was integrated into a
both geometries. Cheney and Zan [19,20] studied the mean surface piloted flight simulation environment. Solution dependent grid
flow topology on a 1:60 scale model of the SFS using oil and pres- adaption was used to refine the mesh in areas of large flow gradi-
sure tappings; off-body flow was examined using smoke visualiza- ents. A comparison of surface pressure coefficients and off-body
tions. Following this, hot-film anemometry was used to investigate velocity components showed reasonable agreement to the NRC
658 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673

data for steady-state computations with the realizable k  e turbu- FLUENT’s pressure-based Navier–Stokes solver was used, with a
lence model. More recently, Yesilel and Edis [24] showed some second-order pressure interpolation scheme. Convective terms
limited improvements to the comparisons presented by Roper, were discretised using a third-order Monotone Upstream-centered
through the use of unsteady simulations with CFX and FLUENT. Fi- Schemes for Conservation Laws (MUSCL) scheme [34] consisting of
nally, Syms [25] used the lattice-Boltzmann technique to perform a blended central differencing/second-order upwind formulation.
time-accurate solutions of SFS and SFS2 airwakes. Good agreement Time integration was performed implicitly using a second-order
to the experimental data is shown, despite a slight over-prediction accurate scheme with dual time stepping.
of RMS turbulence.
This paper reports a significant step forward in DI research by 2.2. Solution strategy
presenting the results of time-accurate CFD computations of the
SFS2 airwake, computed using Detached-Eddy Simulation (DES). Solutions were iterated to steady-state before the unsteady sol-
DES is a relatively new approach to turbulence modelling which ver was activated. A baseline time-step of Dt ¼ 1:88  102 (nor-
is promising for ship airwake applications due to its ability to malised by freestream velocity and ship beam) was chosen based
explicitly resolve turbulent structures for massively separated high on guidelines given in [35] and through comparisons with non-
Reynolds number flows around bluff bodies [26–28]. Its behavior is dimensionalised time-steps used in other DES studies [36,37].
similar to Large-Eddy Simulation (LES), but is computationally Computations were also performed using time-steps scaled by a
cheaper; closer to unsteady Reynolds-averaged Navier–Stokes (UR- factor of 2 in each direction to test solution sensitivity. Comparison
ANS) in terms of required CPU time. A description of the solution of mean flow statistics to experimental data showed that changing
strategy will be given later in the paper and this will be accompa- the time-step had very little effect on the solution. Spectral analy-
nied by a detailed comparison of the CFD results with the NRC sis of velocity fluctuations over the flight deck showed that smaller
wind tunnel data. The ship airwake generation technique de- time-steps were able to resolve progressively more energy at fre-
scribed has also been applied to a Royal Navy Type 23 Frigate; quencies above 10 Hz. However, at full-scale the majority of turbu-
some limited validation of these airwakes against full-scale ane- lent energy in the airwake is known to be in the range 0.1–1 Hz and
mometer data will be presented. An analysis of the SFS2 and Type it is known that disturbances at frequencies above 2 Hz have little
23 Frigate airwake flow topologies will be performed, highlighting effect on pilot workload [3], therefore the increased computational
the dominant flow features which contribute to pilot workload expense of the smaller time-step is not justified.
during the ship landing task. An exercise was performed to test the optimal number of sub-
The main purpose of the current study is the generation of time- iterations; this showed that 10 iterations per time-step gave at
accurate ship airwake data for use in the University of Liverpool’s least two orders of magnitude drop in the continuity residual
HELIFLIGHT simulation environment [29]. Flight trials using this and at least three in the others. Increasing the number of sub-iter-
enhanced airwake model have been performed and pilot com- ations beyond 10 did not increase convergence significantly and
ments are encouraging. Some preliminary results have already added considerably to the required run time.
been obtained [30,31] and a comprehensive analysis of recent sim- Approximately 23 time units were computed to remove tran-
ulation trial data in terms of pilot control activity and workload sients before unsteady sampling began. Flow statistics were then
ratings will be presented in due course. averaged over the next 90 time units. A complete CFD run con-
sisted of 6000 time-steps, with 4800 used for sampling. Computa-
tions were performed to match the experimental wind directions
2. Computational details which are defined using naval terminology, with winds from star-
board denoted as ‘Green’ and winds from port as ‘Red’. Conditions
2.1. Numerical method were therefore computed for a headwind, Green 45° and Green
90°; each run taking approximately 200 h of wall clock time on
Computations were performed using the commercially avail- 32 processors of the University of Liverpool’s high performance
able finite-volume code FLUENT [32], employing DES with the computing cluster.
shear stress transport (SST) k  x turbulence model for closure.
The DES approach uses a modification to the underlying turbulence 2.3. Grid generation and boundary conditions
model to allow medium to large-scale turbulent structures to be
resolved in regions where the computational grid is fine enough. The SFS2 geometry was placed in the centre of a cylindrical do-
This modification links the turbulent length scale to the local grid main with a radius r ¼ 4:5ls , where ls is the ship length; the depth
spacing, D, such that levels of eddy-viscosity are suppressed where of the domain was set to approximately d ¼ 0:75ls . This configura-
D is small. The term which governs the dissipation of turbulent ki- tion (shown in Fig. 2) was chosen as it allowed the WOD condition
netic energy in the SST k  x model involves the parameter, fb ,
which is a constant equal to 1 in the standard model, but modified
according to the following expression in the DES implementation
[33]:
 
Lt
fb ¼ max ;1 ð1Þ
C DES D

where Lt is the turbulent length scale and C DES is a constant within

the DES model usually given a value of 0.61 (retained in the current
study). In regions of the flow where D is small (i.e. below the local
turbulent length scale), the value of fb is increased above unity,
thus causing an increase in dissipation of turbulent kinetic energy.
It is the resultant decrease in eddy-viscosity which prevents flow
field perturbations from becoming artificially damped by the turbu-
lence model and allows turbulent structures to propagate. Fig. 2. Schematic showing SFS2 grid configuration.
 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673 659

to be altered simply by changing the x and y components of free-

stream velocity. The outer, curved, boundary was specified as a
pressure far-field, both upper and lower surfaces set as walls with
zero shear-stress and the ship surface was modelled as a wall with
a no-slip boundary condition imposed.
The SFS2 is a relatively simple geometry and could easily be
meshed using hexahedral cells on a block-structured grid. How-
ever, the main purpose of these computations was to serve as a val-
idation exercise for DES applied to ship airwakes, with the aim of
applying the same methods to more complex ship geometries.
Ships such as the Type 23 Frigate have superstructures with intri-
cate geometric features which would prove extremely difficult to
mesh using a structured approach. Such geometries are ideally sui-
ted to unstructured meshes as cells can be tightly clustered around
the smaller features and allowed to grow larger away from the ship
where flow gradients are smaller. The SFS2 was therefore meshed Fig. 3. Surface mesh covering the SFS2 and the lower wall boundary.
using a hybrid structured/unstructured mesh to test whether satis-
factory results could be obtained using this approach. Some suc-
cess has already been reported using unstructured grids with When comparing mean flow quantities over the flight deck for
DES [38], as the isotropic nature of tetrahedra suits the method the three grids little difference was found between the results.
well. Fig. 4 shows mean velocity magnitude and turbulence intensity
The ship surface was meshed with triangular elements and 15 along a lateral line above the flight deck, with wind tunnel data
layers of prisms were grown from the surface to resolve the viscous also plotted for reference. It is clear that none of the grids offer sig-
boundary layer. Spacing normal to the wall was chosen to give wall nificant improvement in comparison to experimental data. The
unit values of yþ ¼ O(10) and an expansion ratio of 1.3 was applied. similarity of the three sets of results, despite the difference in cell
The geometry was placed inside an unstructured sub-domain size between them, indicates grid-independent solutions. The
which was meshed using tetrahedral elements. Size functions were coarsest grid, C, would appear to give satisfactory results in terms
applied to ensure smooth cell growth away from the superstruc- of mean quantities, however it is worth re-iterating that the pur-
ture and also to create a refinement region over the flight deck. pose of these computations is to provide accurate unsteady data
This is analogous to Spalart’s ‘Focus Region’ [35], where cells for use in flight simulation. A coarser grid requires a larger time-
should be small and, as far as possible, isotropic in the area of inter- step to maintain an appropriate Courant number; for grid C this
est to fully utilise the power of DES. In regions outside of the sub- would correspond to a solution frequency of 56 Hz. As the output
domain a structured meshing approach was employed. from these computations would be used to produce unsteady air-
To test grid independence, and in accordance with the guide- wake data for use in piloted flight simulations, it was felt that a
lines in [35], three target grid spacings ðD0 Þ were analyzed by higher airwake update frequency than this may be required to en-
choosing a baseline D0 and scaling this up and down by a factor sure an appropriate level of simulator fidelity. The run-time and
pffiffiffi memory requirements for grid B were within the capability of
of 2. The test computations were performed for a headwind at
a freestream speed of 40 kts and the grids were denoted A the compute cluster, therefore it was decided that grid B would
ð10:4  106 cellsÞ, B ð5:8  106 cellsÞ and C ð3:3  106 cellsÞ for be used for all subsequent computations to ensure that spatial
spacings of D0 =h ¼ 4:12  102 ; 5:83  102 and 8:33  102 , and temporal resolutions were adequate.
respectively, where h is hangar height. The surface mesh for spac-
ing B can be seen in Fig. 3. Time-steps for grids A and C were scaled 3. Results and comparison
linearly with D0 to ensure that temporal and spatial resolution re-
mained well-balanced. The requirement for a reduction in time- In this section results from the CFD computations will be com-
step with refinement of the grid adds further to the computational pared with wind tunnel data to validate the CFD methodology.
time required for runs on the finer grids. Both qualitative and quantitative comparisons will be made of

Fig. 4. Comparison of velocity magnitude (a) and turbulence intensity (b) on a lateral line over the flight deck for computational grids A, B and C. Results are normalised by
freestream velocity and ship beam.
660 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673

time-averaged flow quantities and velocity spectra at several WOD mounted on a turntable and boundary layer suction was employed
conditions. All velocity data is normalised with respect to the free- to ensure a uniform incident velocity profile. Fig. 6 shows the scale
stream velocity magnitude. Distances are defined relative to the model of the SFS2 mounted inside the wind tunnel.
airwake origin (located at the base of the hangar on the ship centr-
eline) with longitudinal, lateral and vertical locations normalised
by deck length, ship beam and hangar height, respectively. The axis 3.2. Reynolds number dependence
system is defined such that x is positive to stern, y is positive to
starboard and z is positive up. The SFS2 geometry is a bluff-body, consisting of rectangular
surfaces and sharp edges. It is generally assumed that the flow over
such structures is insensitive to Reynolds number, so the flow
3.1. Experimental details topology at full scale should be replicated at model scale. Conse-
quently, all CFD runs were performed at full scale, equivalent to
The experiments were conducted in the 2 m  3 m low-speed a Reynolds number of 2:26  107 where ship beam, b, is used as
wind tunnel at the Aerodynamics Laboratory of the National Re- the characteristic length. However, to test Reynolds number sensi-
search Council, Canada. Hot-film anemometry was used to obtain tivity a single headwind case was computed to match the wind
u–v and u–w data consisting of mean velocities and turbulence tunnel conditions, at a Reynolds number of 6:58  105 . It was
intensities along a series of experimental maps over the SFS2 found that the flow patterns were essentially the same, with some
(Fig. 5). Unsteady velocity spectra were also recorded at several slight differences in the location of features such as reattachment
points over the superstructure. The 1:100 scale model was points and vortex cores. This is illustrated in Fig. 7, where the cen-

Fig. 5. Location of experimental maps during wind tunnel testing. Black dots indicate spectra recording points. Distance along deck measured from the hangar given in
percent.

Fig. 6. SFS2 model mounted inside the NRC 2 m  3 m low-speed wind tunnel (image courtesy of NRC, reprinted with permission).
 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673 661

Fig. 7. Location of recirculation centres (circles) and reattachment points (bold arrows) for CFD calculations at model scale and full scale.

tres of the recirculation bubbles are indicated; the model scale case approximately 5% higher than the other two components over
being below and slightly aft the location of the full scale case. Sim- the flight deck. This is not seen from the CFD results, where the
ilarly, flow is shown to reattach to the deck slightly earlier at model x-component is only slightly higher. Overall though, the agreement
scale, at 45% of deck length compared to 49% at full scale. These ef- for the headwind case is very good, with turbulence intensity lev-
fects are not anticipated to have a significant impact on the data els matched well, particularly in the lateral and vertical directions.
comparison exercise, although subtle differences in the location As the wind direction changes from a headwind around to a
of dominant flow features such as vortices or shear layers can af- Green 45° condition (wind at 45° from starboard), the flow pattern
fect agreement when plotting at a limited number of data points. over the deck is altered dramatically. Weak vortical structures are
seen to appear over the superstructure as the wind moves off the
3.3. Comparison to experimental data bow, before strengthening considerably and impacting the flight
deck as the wind reaches Green 30° [19]. At Green 45°, flow over
Data plots have been drawn along a lateral line located at ‘Map the flight deck is dominated by separation from the windward deck
1c/3c’ in Fig. 8 for comparison. The line is located at 50% of the edge and a vortical structure is formed from the upper corner of
flight deck length, at hangar height, and spans the equivalent of the windward edge of the hangar. These conditions prove challeng-
two beam widths. Many such plots have been studied and show ing for numerical methods to predict as the off-body flow patterns
similar trends; this location was chosen as it represents the region are sensitive to the shear layer separation angles. Although separa-
closest to where a helicopter would be hovering during a deck tion points are generally fixed by the sharp edges, strong stream-
landing. line curvature in the separating shear layer is difficult for eddy-
Fig. 8 shows mean velocity components and turbulence intensi- viscosity based turbulence models to capture and this causes diffi-
ties for the headwind case. A reduction in longitudinal velocity can culty in obtaining accurate predictions of shear dominated flow
be seen towards the centreline, indicating that the plotting location trajectories [39].
intersects the wake behind the hangar. A corresponding downdraft A comparison between CFD and wind tunnel results for a Green
is also seen, showing that the flow separated from the hangar roof 45° WOD condition is shown in Fig. 9. Velocity components are
is heading down towards the deck. The slight asymmetry shown in compared in Fig. 9 and, while qualitatively the trends are well cap-
the experimental results and as computed by Syms [25] is not evi- tured by the CFD, there are some obvious differences in both the u
dent in the present computations. However, all of the trends exhib- and v components. A similar level of disagreement in the lateral
ited in the wind tunnel data are replicated by the CFD and, in most velocity component is also seen in Syms’ computations [25]. Inter-
locations, excellent agreement is shown. Studying the turbulence estingly, disagreement is seen all the way out to y/b = 1, indicating
data in Fig. 8, again the trends are in agreement although there that flow is approaching the ship from slightly different directions
is some discrepancy, particularly in the longitudinal component. between the CFD and wind tunnel. Indeed, when velocity vectors
The experimental data shows that the longitudinal turbulence is are plotted, as shown in Fig. 10, it is clear that the incident flow

Fig. 8. Headwind mean velocities (a) and turbulence intensities (b) normalised by V 1 at 50% deck length, plotted at hangar height. Lateral position normalised by ship beam.
662 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673

Fig. 9. Green 45° mean velocities (a) and turbulence intensities (b) normalised by V 1 at 50% deck length, plotted at hangar height. Lateral position normalised by ship beam.

Fig. 10. Velocity vectors comparing CFD and experimental results for the Green 45° case plotted on a y–z plane at x/l = 0.5 (a), an x–z plane at y/b = 0 (b) and an x–y plane at
z/h = 1 (c).

in the wind tunnel is aligned approximately 5° closer to the ship low velocity flow. As the rear of the deck is approached the shear
centreline than the CFD calculations. It appears that the upstream layer moves progressively to port and becomes thicker, although
flow in the wind tunnel is affected more strongly by the presence it is slightly closer to starboard in the wind tunnel case, for reasons
of the ship than in the CFD; this reduces the effective WOD angle described above. The velocity deficit behind the hangar is more
causing two main effects. Firstly, the u and v velocity components pronounced in the CFD results; however the relatively coarse spa-
over the starboard side of the flight deck are higher and lower, tial sampling taken in the wind tunnel may mean that local low-
respectively, than the CFD results (as seen in Fig. 9). Secondly, velocity regions are missed.
the shear layer separating from the windward hangar edge has a Only a limited experimental data-set consisting of v and w com-
shallower separation angle in the wind tunnel, meaning that its ef- ponents was available for the Green 90° WOD condition, so com-
fects are seen closer to the starboard deck edge than in the CFD. parisons are restricted to contour plots. Contours of vertical
This can be seen by the difference in locations of the lateral turbu- velocity and the vertical component of turbulence on Map 6a are
lence intensity peaks in Fig. 9. These peaks are caused by flapping plotted in Fig. 12. Good agreement is shown, with an increase in
of the shear layer, and while the CFD peak is at the approximate the vertical velocity component observed close to the starboard
location y/b = 0.16, the wind tunnel peak is seen at y/b = 0. As edge of the superstructure due to flow separation. A corresponding
with the headwind case, the longitudinal turbulence component region of elevated turbulence is seen close to the superstructure
is under-predicted by the CFD. edge, with increasing intensity towards the base of the funnel.
Despite the differences identified, the flow field computed by Having performed analysis at headwind, Green 45° and Green
CFD is qualitatively very similar to the wind tunnel data. Fig. 11 90° WOD conditions it has been shown that the DES-based CFD ap-
shows contours of velocity magnitude and turbulence intensity proach is able to predict the SFS2 airwake with good accuracy. De-
plotted on experimental Maps 3a–3c. The location of the hangar spite slight differences in the location and orientation of separating
edge shear layer can be identified by the transition from high to shear layers, the magnitude of the unsteadiness in the flow over
 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673 663

Fig. 11. Contours of velocity magnitude (a) and turbulence intensity (b) for the Green 45° case.

Fig. 12. Contours of vertical velocity (a) and turbulence intensity (b) for the Green 90° case.

the flight deck is well captured; which is of prime importance in scaled using the Strouhal number to match the full-scale CFD con-
creating a realistic ship airwake model for flight simulation. ditions. Spectral characteristics are extracted from the CFD data
using a fast Fourier transfrom (FFT) algorithm.
3.4. Spectral characteristics At the spectra point on Map 1c (located at a distance
x=l ¼ 0:5; y=b ¼ 0:4 and a height of z=h ¼ 0:75 above the deck)
Plots of Power Spectral Density (PSD) are shown in Figs. 13 and agreement between CFD and wind tunnel data is very good, both
14, where velocity data has been recorded at points on Map 1c and in terms of frequency content and power. The longitudinal velocity
Map 2a (see Fig. 5), respectively. The experimental data has been component (Fig. 13a) exhibits a gradual drop-off in the range 0.2–

Fig. 13. Power Spectral Density plots of longitudinal (a) and lateral (b) velocity components recorded at Map 1c spectra point.
664 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673

imately St = 0.12, using funnel width as the characteristic length.

This is similar to the values reported by other researchers for
rectangular cylinders in cross-flow [40–42]. However, compared
to these studies, the turbulence levels incident on the funnel
and the Reynolds number are much higher in the present ship
airwake computations; the fact that the funnel is truncated
rather than of infinite span will also undoubtedly affect the sep-
aration characteristics, thus impacting the shedding frequency. It
is encouraging to see that the CFD technique is able to pick up
the shedding and match the wind tunnel results well, despite
the fact that the computational mesh is not optimised for de-
tailed analysis of the funnel wake.

4. Ship airwake flow topology

This section will identify the mechanisms responsible for the

generation of large-scale turbulent structures over the flight deck
Fig. 14. Power Spectral Density plot of the lateral velocity component recorded at and highlight the dominant flow features which are expected to
Map 2a spectra point. impact upon helicopter–ship operations.

4.1. Headwind
0.3 Hz, with the CFD results matching the experimental gradient
very well despite a small power deficit in the CFD data between At the headwind condition flow separates from the front edge
0.7 and 0.9 Hz. The lateral component (Fig. 13b) shows a steady in- of the superstructure, resulting in high levels of shear and the
crease in power to a small peak at 0.9 Hz before dropping off; again formation of turbulent eddies which are convected downstream
the CFD matches both characteristics well. towards the funnel. Figs. 15–17 illustrate this clearly, in particu-
The spectra point on Map 2a is located behind the funnel, di- lar Fig. 17 which uses the k2 criterion [43] to identify the loca-
rectly above the hangar on the ship centreline at a height tion of vortex cores (only contours with a value k2 6 0 are
z=h ¼ 1:38 above the deck. Fig. 14 shows a definite peak in the plotted). The value k2 is the second eigenvalue of S2 þ X2 , where
lateral PSD plot at 0.8–0.9 Hz indicating the possibility of weak S and X are the symmetric and anti-symmetric parts of the
shedding from the funnel which, in effect, is a truncated rectan- velocity gradient tensor, respectively. Values of k2 below zero
gular cylinder. This corresponds to a Strouhal number of approx- show the presence of a vortex core, with increasingly negative

Fig. 15. Contours of mean (top) and instantaneous (bottom) velocity magnitude for a headwind, plotted on a plane at z=h ¼ 1:15 above the deck.
 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673 665

Fig. 16. Contours of mean (top) and instantaneous (bottom) velocity magnitude for a headwind, plotted on a plane at y=b ¼ 0.

Fig. 17. Instantaneous contours of k2 indicating the presence of vortex cores for a headwind, plotted on a plane at z=h ¼ 1:15 (top) and y=b ¼ 0 (bottom).

values indicating stronger vortices. It is also worth noting the evidence to suggest that this structure exhibits coherent vortex
significant differences between the mean and instantaneous shedding. At high hover over the flight deck a helicopter would
flow-fields; further evidence of the highly unsteady nature of be above the hangar roof and directly behind the funnel, there-
the flow. The separated region re-attaches to the superstructure fore the turbulent structures shed from the funnel would be
at approximately 3 hangar heights downstream of the front likely to adversely affect handling qualities in this situation.
edge, although the exact location is transient due to flapping As flow approaches the flight deck it separates from the
of the re-attaching shear layer. superstructure at the top and sides of the hangar, forming a
The strength of the vortical structures appears to diminish as 3D recirculation bubble in the lee of the hangar. The separation
they travel down the superstructure, but the flow is then re- is less severe than that experienced at the front edge of the
energised as it encounters the funnel. Fig. 17 shows the turbu- superstructure, resulting in weaker shear layers and a corre-
lent eddies shed from the sharp edges of the rectangular funnel sponding reduction in the strength of turbulent eddies produced
and the instantaneous contours of Fig. 15 show the resultant by this separation. Nonetheless, the instabilities in these separat-
asymmetric wake in the lee. As discussed in Section 3.4, this is ing flows is the dominant mechanism for turbulence generation
666 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673

Fig. 18. Iso-surfaces of k2 ¼ 0 indicating the location of vortex cores over the superstructure for a Green 45° wind.

are plotted on a series of y-z slices. Again, the mean and instanta-
neous flow-fields show significant differences; high strength vorti-
cal structures are present throughout the wake in Fig. 20, but
largely disappear in the average. This highlights the complex and
chaotic nature of the separated flow over the superstructure. The
secondary vortex located to the starboard side of the funnel is pres-
ent in both diagrams, indicating that it is a relatively stationary
structure.
Fig. 21 shows iso-surfaces of k2 over the flight deck, with three
distinct flow features highlighted. Feature (a) is the starboard
deck-edge vortex, formed as flow travelling along the lower star-
board side of the superstructure meets freestream flow at the star-
board deck edge. The top of this structure is roughly in line with
the hangar, reaching its maximum height in the middle of the flight
deck. Feature (b) is the starboard hangar edge shear layer, which is
formed as the flow attached along the starboard edge of the super-
Fig. 19. Smoke flow at Green 45° observed during NRC wind tunnel tests of the structure separates from the vertical edge of the hangar. In CFD
SFS1 geometry, indicating superstructure vortices (image from Ref. [19], reprinted
animations this feature is seen to be unsteady in nature, with a
with permission).
flapping motion commonly seen in bluff-body separation. This is
a major contributing factor to turbulence over the flight deck, par-
ticularly in regions below the hangar. Feature (c) is the secondary
over the flight deck, in addition to the coherent structures shed superstructure vortex identified above, which appears to break up
from the funnel. and shed helical vortical structures as it encounters the flight deck.
This flow feature is responsible for most of the turbulence encoun-
tered over the flight deck at locations above hangar height and
4.2. Green 45° wind
would therefore pose significant difficulties to rotorcraft operating
to and from such a ship.
Flow features at Green 45° are significantly different to the
The interactions between the flow features identified above re-
headwind condition. As the freestream encounters the starboard
sult in a highly complex, unsteady flow-field over the flight deck.
corner of the superstructure two vortices are formed (Fig. 18);
Compared to the turbulent structures observed for the headwind,
one aligned laterally with the front edge of the superstructure,
the Green 45° condition produces larger vortical structures with
the other oriented longitudinally along the windward superstruc-
higher strength (Fig. 22). The impact of this is likely to be turbulent
ture edge which then widens as it approaches the funnel. This flow
fluctuations with larger spatial scales and higher velocities. This is
pattern was also observed by Cheney and Zan1 [19] as shown in
confirmed through a comparison of Figs. 8 and 9, which show that
Fig. 19.
turbulence intensities over the flight deck peak at approximately
The longitudinal vortex is broken up as it passes over the funnel,
20% for the headwind as opposed to 30% for a Green 45° wind.
but the flow passing up and over the starboard side of the super-
structure causes a secondary vortex to form between the longitu-
dinal superstructure edge and the funnel. This is illustrated in 5. Type 23 Frigate
Fig. 20, where instantaneous and time-averaged contours of k2
Following successful validation of the CFD approach with the
1
Although Cheney and Zan studied the SFS1 which has no pointed bow and a
SFS2 geometry, the same methodology was applied to a Type 23
shorter superstructure (Fig. 1), for qualitative comparisons the SFS1 and SFS2 Frigate (T23), which is a class of warship currently in service with
geometries can be expected to exhibit similar flow features. the UK Royal Navy. A picture of the T23 is shown in Fig. 23 and the
 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673 667

Fig. 20. Contours of k2 indicating the presence of vortex cores for a Green 45° wind; instantaneous (a) and time-averaged (b).

Fig. 21. Iso-surfaces of k2 ¼ 0 indicating the location of instantaneous vortex cores over the flight deck for a Green 45° wind. Labelled flow structures: deck edge vortex (a);
hangar edge shear layer (b); and superstructure edge vortex (c). Outline of SFS2 shown by dashed line.

Fig. 22. Instantaneous contours of k2 indicating the presence of vortex cores for a Green 45° wind, plotted on a plane at z=h ¼ 1:15 (top) and y=b ¼ 0 (bottom).
668 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673

CAD geometry used is shown in Fig. 24. The geometry is clearly far
more complex than the SFS2 and part of the motivation for the
study was to determine whether this increase in geometric fidelity
resulted in a noticeable difference in airwake topology. It was
anticipated that the large ‘notch’ in the starboard edge of the T23
hangar may, for certain Green winds, generate flow features not
seen on the SFS2. Experimental data was available for validation
of the T23 airwakes, in the form of full-scale anemometer readings
taken at sea.

5.1. Experimental details

Full-scale at-sea testing of the T23 airwake were performed by

DRA (now QinetiQ) on board HMS Iron Duke in 1994. Mean veloc- Fig. 25. Anemometer locations during at-sea Type 23 airwake testing.
ities and turbulence data behind the hangar was obtained using 3-
axis propeller anemometers mounted on a mast at heights of
approximately z=h ¼ 0:5, 1.0 and 1.5 above the deck. Longitudinal Due to the fact that the full-scale experimental data was ob-
and lateral anemometer locations are shown in Fig. 25. It is known tained at sea, it was necessary to model the effects of the earth’s
that such anemometers can have a significant in-built inertia and atmospheric boundary layer (ABL) during the T23 CFD computa-
suffer attenuation at frequencies above 0.2 Hz [44], so this should tions. The ABL causes a velocity reduction near the sea surface
be considered when comparing turbulent quantities. Data were which can be modelled using the following power law [45]:
made available at Green 10°, Green 30° and Green 90° WOD condi-  a
tions. Only comparisons at Green 10° will be presented here for z
U ¼ U ref ð2Þ
brevity, although computations were also performed at headwind zref
and Green 45° so that qualitative comparisons to the SFS2 airwakes
where U ref is the velocity at the reference height, zref , and a is a con-
can be made.
stant which depends on the surface terrain. In the current study val-
ues of U ref ¼ 35 m=s and zref ¼ 300 m were used, giving a wind
5.2. Computational details speed of approximately 50 kts at the nominal T23 anemometer
height. The constant, a, was set to 0.13 as recommended by Couni-
As the overall dimensions of the T23 are very similar to the han [45] for an ocean surface. During the present study the mod-
SFS2, all numerical details (grid spacing, time-step) were retained elled effects of the ABL were limited to an appropriate velocity
from the SFS2 computations. Due to the increased geometric com- profile; the effects of increased freestream turbulence were not in-
plexity, the cell count was higher than the SFS2 grid at 7:4  106 cluded. It was decided that runs would be performed with and
cells. Computations were performed in exactly the same way as without the effect of the ABL to determine what, if any, impact this
for the SFS2, as described in Section 2.2. had on the nature of the airwake. Polsky [46] has shown that for
beam winds (90° from port or starboard) the inclusion of the ABL
can significantly improve agreement with full-scale experimental
data.

5.3. Results

Figs. 26 and 27 show a comparison of velocity components and

turbulence intensities for the Green 10° WOD condition. Map 1a is
closest to the hangar and Map 1c is furthest aft. Due to the small
number of experimental locations from the full-scale data it is dif-
ficult to draw definitive conclusions on agreement between data-
sets. However, it is clear that the trends shown by the at-sea data
are well matched by CFD. This is all the more remarkable given the
fact that the CFD geometric model is simplified, lacking many of
the small-scale features of the full sized ship (masts, netting, an-
Fig. 23. A Royal Navy Type 23 Frigate.
tenna arrays).
The two CFD computations predict essentially the same wake
pattern, with the ‘ABL’ velocities a factor of approximately 0.8 low-
er than the ‘no-ABL’ results. This is due to the fact that results are
normalised by the velocity at anemometer height. For a uniform
inflow profile (i.e. the no-ABL case) at a given freestream condition
the velocity over the flight deck will be the same as the anemom-
eter velocity, but where the ABL is included flow velocities over the
flight deck will be reduced by an amount dependent on the vertical
distance between the deck and the anemometer. It is interesting to
note, however, that the full-scale results show normalised longitu-
dinal velocities approaching unity at the windward edge of the
deck, despite these locations being significantly below the ane-
mometer. It is entirely possible that the wind conditions on the
Fig. 24. Type 23 Frigate CAD geometry used for current study. day of the at-sea testing differed from the generic power-law
 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673 669

Fig. 26. Longitudinal (a), lateral (b) and vertical (c) velocity components in the Type 23 Frigate airwake plotted at hangar height for a Green 10° condition.

ABL model used within the CFD computations. Air flow over the airwake are sensitive to such boundary conditions. Therefore, it is
flight deck is a result of ship forward speed, prevailing wind, or a recommended that ship airwake computations for flight simulation
combination of both. Ship motion causes uniform flow, whereas purposes should include an appropriate ABL velocity profile in order
the wind is subject to an ABL profile; the vector sum of both com- to improve realism.
ponents will change the effective power-law exponent. It is there-
fore expected that this will account for a certain amount of 5.4. Flow topology
variance between CFD and at-sea results.
The turbulence data follows a similar pattern; the intensity is Comparing Figs. 16 and 28 it can be seen that the headwind
slightly lower for the ‘ABL’ case in most locations due to the lower condition displays very similar characteristics to the SFS2 flow-
incident velocities. A corresponding improvement in agreement field over the flight deck. The main effect of the geometric features
with full-scale data is found, consistent with Polsky’s findings [46]. on the superstructure is to retard the flow so that velocities around
This reduction in airwake turbulence for the ‘ABL’ computations is the flight deck are slightly lower than seen on the SFS2. Neverthe-
an important finding, as it shows that turbulent fluctuations in the less, the mean and instantaneous velocity fields shown in Fig. 28
670 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673

Fig. 27. Longitudinal (a), lateral (b) and vertical (c) turbulence intensities in the Type 23 Frigate airwake, plotted at hangar height for a Green 10° condition.

exhibit the same large-scale flow features as seen for the SFS2. One features can give rise to differences in large-scale stationary (in a
difference worth noting is the effect of the large mast to the fore of time-averaged sense) flow patterns over the flight deck. It should
the T23, which appears to shed vortical structures. These propa- be emphasised that in both cases these features are highly unstea-
gate downstream, but are at a height unlikely to affect helicopter dy in nature, with both causing the production of turbulence over
operations. the deck. However, it is known that both mean and fluctuating
There are some clear differences between the T23 and SFS2 air- velocities over the deck can contribute to pilot control strategy
wakes at Green 45°. Fig. 29 shows the effect of the windward han- during landing [31], so it is important that large-scale mean flow
gar notch on the mean flow pattern over the deck. Flow curves features are captured.
smoothly around the vertical hangar edge of the SFS2, whereas Another difference between the T23 and SFS2 airwakes at Green
on the T23 the presence of the notch causes the formation of a vor- 45° is that there are fewer identifiable, coherent flow features over
tex which is aligned more longitudinally with the deck than the the flight deck in the T23 case. This is in part due to the hangar
freestream. This suggests that relatively small-scale geometric being set back from the deck edge, resulting in a deck edge vortex
 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673 671

Fig. 28. Contours of mean (top) and instantaneous (bottom) velocity magnitude for a headwind, plotted on a plane at y=b ¼ 0.

which is less pronounced. The staggered vertical hangar edge also

appears to lessen the intensity of the separating shear layer.
A comparison of Figs. 22 and 30 shows that turbulent structures
over the deck are smaller for the T23, although their strength is
comparable to the SFS2 case. The dominant cause of turbulence
generation at this WOD condition is separation from the windward
corner of the hangar; these vortical structures appear to be shed
more uniformly than for the SFS2, possibly due to less interaction
with other large-scale flow features. Fig. 30 also shows that high-
strength vortical structures are concentrated more on the port side
of the deck for the T23, as opposed to the SFS2 case (Fig. 22) where
vorticity occupies a larger proportion of the flight deck.

6. Summary and concluding remarks

Unsteady computations of the flow over the SFS2 and T23 ships
have been performed using the commercial CFD code, FLUENT. The
use of DES for turbulent closure is shown to be suitable for such
flows due to its ability to capture the large-scale turbulent struc-
tures shed from ship superstructures. Unstructured grids were
used to enable the close control of cell size and to simplify the pro-
cess of meshing complex ship geometries; this approach has
proved successful in these respects.
A comprehensive validation exercise, comparing CFD results for
the SFS2 to wind tunnel data for three WOD conditions has been
described. Mean velocity and turbulence data have been compared
using line plots, contours and vectors, showing good agreement in
most cases. An analysis of PSD plots showed that the computa-
tional method was able to generate levels of turbulent power com-
parable to the wind tunnel, and also matched the frequency roll-off
over the relevant bandwidth. This is essential if CFD generated air-
wake data is to be used for piloted flight simulation of deck landing
tasks.
Airwake data for the T23 was compared to full-scale experi- Fig. 29. Mean pathlines for the Green 45° condition, indicating the separation
mental data, showing good agreement with trends observed at characteristics from the windward hangar edge of the SFS2 (a) and the Type 23
sea. Computations of airwakes with and without the influence of Frigate (b). Pathlines coloured by vorticity magnitude.
672 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673

Fig. 30. Instantaneous contours of k2 indicating the presence of vortex cores for the Green 45 condition, plotted on a plane at z=h ¼ 1:15 (top) and y=b ¼ 0 (bottom).

an ABL showed that the inclusion of a representative ABL velocity Type 23 Frigate validation data was provided by the UK Defence
profile caused lower velocities and a corresponding reduction in Science and Technology Laboratory (DSTL).
turbulence over the deck. This improved agreement with full-scale
data, particularly in terms of turbulence intensity. References
Post-processing of the CFD data allowed extraction of the main
topological flow features for winds from ahead and Green 45°. It [1] Carico D, Fang R, Finch RS, Geyer Jr WP, Krijns HW, Long K. Helicopter/ship
qualification testing. Technical Report. RTO-AG-300-V22, RTO/NATO; 2003.
was shown that shear layer separation and vortex formation from [2] Advani SK, Wilkinson CH. Dynamic interface modelling and simulation – a
sharp edges is the dominant mechanism for turbulence generation unique challenge. In: Royal aeronautical society conference on helicopter flight
over the flight deck. Significant differences exist between head- simulation, London, UK; 2001.
[3] Zan SJ. On aerodynamic modelling and simulation of the dynamic interface.
wind and Green 45° airwakes, with the latter containing turbu-
Proc Inst Mech Eng, Part G: J Aerospace Eng 2005;219(5):393–410.
lence with larger amplitude and spatial scales. [4] Bogstad MC, Habashi WG, Akel I, Ait-Ali-Yahia D, Giannias N, Longo V.
Differences between the SFS2 and T23 airwakes have been Computational-fluid-dynamics based advanced ship-airwake database for
found for the WOD conditions studied. The T23 airwake is shown helicopter flight simulators. J Aircraft 2002;39(5):830–8.
[5] Roper DM, Owen I, Padfield GD, Hodge SJ. Integrating CFD and piloted
to contain smaller turbulent structures with similar strength to simulation to quantify ship–helicopter operating limits. Aeronaut J
the SFS2 and less complex interactions between large-scale vorti- 2006;110(1109):419–28.
cal features are observed. Larger differences exist in terms of the [6] Zan S. Technical comment on ‘computational-fluid-dynamics based advanced
ship-airwake database for helicopter flight simulation’. J Aircraft
mean flow pattern over the deck at Green 45°, due to the presence 2003;40(5):1007.
of the T23 windward hangar ‘notch’. It is still unclear whether [7] Roper DM, Owen I, Padfield GD. CFD investigation of the helicopter–ship
these differences are significant enough to be detected by a pilot dynamic interface. In: 61st AHS annual forum proceedings, vol. 2, Grapevine,
TX; 2005. p. 1985–2002.
during simulated deck trials, indeed the use of a simplified ship [8] Polsky SA, Bruner CWS. Time-accurate computational simulations of an LHA
geometry may be adequate for many purposes. However, it has ship airwake. In: 18th AIAA applied aerodynamics conference, Denver, CO;
been shown that the intelligent use of unstructured meshing tech- 2000.
[9] Polsky SA. A computational study of unsteady ship airwake. In: 40th AIAA
niques for complex ship geometries can give an increase in airwake aerospace sciences meeting, Reno, NV; 2002.
fidelity at a modest increase in computational cost. The use of real- [10] Bunnell JW. An integrated time-varying airwake in a UH-60 Black Hawk
istic geometries for use in CFD airwake simulations is, therefore, shipboard landing simulation. In: AIAA modeling and simulation technologies
conference and exhibit, Montreal, Canada; 2001.
recommended.
[11] Roscoe MF, Thompson JH. JSHIP’s dynamic interface modeling and
simulation system: a simulation of the UH-60A helicopter/LHA
shipboard environment task. In: American helicopter society 59th
annual forum, Phoenix, AZ; 2003.
Acknowledgements [12] Lee D, Sezer-Uzol N, Horn JF, Long LN. Simulation of helicopter shipboard
launch and recovery with time-accurate airwakes. J Aircraft
2005;42(2):448–61.
The first author is funded by an EPSRC Doctoral Training Award [13] Lee D, Horn JF. Simulation of pilot workload for a helicopter operating in a
and by Westland Helicopters Ltd. ANSYS Inc have been most gen- turbulent ship airwake. Proc Inst Mech Eng, Part G: J Aerospace Eng
2005;219(5):445–58.
erous in their assistance. The authors would like to thank Mr Cliff
[14] Lee D, Horn JF, Sezer-Uzol N, Long LN. Simulation of pilot control activity
Addison for invaluable help with compute cluster configuration during helicopter shipboard operation. In: AIAA atmospheric flight mechanics
and Mr Steven Hodge (BAE Systems) for providing the Type 23 conference and exhibit, Austin, TX; 2003.
Frigate CAD model. The SFS2 validation data was derived by the [15] McKillip RM, Boschitsch A, Quackenbush T, Keller J, Wachspress D. Dynamic
interface simulation using a coupled vortex-based ship airwake and rotor
National Research Council Canada, and provided under the aus- wake model. In: American helicopter society 58th annual forum, Montreal,
pices of The Technical Co-operation Program (TTCP). The full-scale Canada; 2002.
 J.S. Forrest, I. Owen / Computers & Fluids 39 (2010) 656–673 673

[16] Guillot MJ. Computational simulation of the air wake over a naval transport American helicopter society 64th annual forum, vol. 1. Montreal, Canada;
vessel. AIAA J 2002;40(10):2130–3. 2008. p. 339–51.
[17] Tai TC, Carico D. Simulation of DD-963 ship airwake by Navier–Stokes method. [31] Forrest JS, Hodge SJ, Owen I, Padfield GD. An investigation of ship airwake
J Aircraft 1995;32(6):1399–401. phenomena using time-accurate CFD and piloted helicopter flight simulation.
[18] Wilkinson C, Zan S, Gilbert N, Funk J. Modelling and simulation of ship air In: 34th European rotorcraft forum, Liverpool, UK; 2008.
wakes for helicopter operations – a collaborative venture. In: AGARD [32] Fluent, Ver. 6.3.26. Lebanon, NH: Fluent Inc.; 2006.
symposium on fluid dynamics, problems of vehicles operating near or in the [33] Menter F, Kuntz M, Langtry R. Ten years of industrial experience with the SST
air–sea interface, Amsterdam, Netherlands; 1998. turbulence model. In: Hanjalic K, Nagano Y, Tummers M, editors. Turbulence,
[19] Cheney BT, Zan SJ. CFD code validation data and flow topology for the technical Heat and Mass Transfer, vol. 4. Redding (CT): Begell House; 2003. p. 625–32.
co-operation program AER-TP2 simple frigate shape. Technical Report. LTR-A- [34] VanLeer B. Toward the ultimate conservative difference scheme. IV. A second
035, NRC–CNRC; April 1999. order sequel to Godunov’s method. J Computat Phys 1979;32:101–36.
[20] Zan SJ. Surface flow topology for a simple frigate shape. Can Aeronaut Space J [35] Spalart PR. Young-person’s guide to detached-eddy simulation grids. Technical
2001;47(1):33–43. Report. NASA/CR-2001-211032, NASA; 2001.
[21] Reddy KR, Toffoletto R, Jones KRW. Numerical simulation of ship airwake. [36] Hedges LS, Travin AK, Spalart PR. Detached-eddy simulations over a simplified
Comput Fluids 2000;29(4):451–65. landing gear. J Fluids Eng, Trans ASME 2002;124(2):413–23.
[22] Liu J, Long LN. Higher order accurate ship airwake predictions for the [37] Forsythe JR, Squires KD, Wurtzler KE, Spalart PR. Detached-eddy simulation of
helicopter/ship interface problem. In: Annual forum proceedings – American the F-15E at high alpha. J Aircraft 2004;41(2):193–200.
helicopter society, vol. 1 of proceedings of the 54th annual forum – part 2 (of [38] Morton SA, Forsythe JR, Squires KD, Wurtzler KE. Assessment of unstructured
2), Washington, DC; 1998. p. 58–70. grids for detached eddy simulation of high Reynolds number separated flows.
[23] Wakefield NH, Newman SJ, Wilson PA. Helicopter flight around a ship’s In: Proceedings of the 8th international conference on numerical grid
superstructure. Proc Inst Mech Eng, Part G: J Aerospace Eng 2002;216(1):13–28. generation for computational field simulations, Honolulu, HI; 2002. p. 571–86.
[24] Yesilel H, Edis FO. Ship airwake analysis by CFD methods. Am Inst Phys Conf [39] Cheng GC, Farokhi S. On turbulent flows dominated by curvature effects. J
Proc 2007;936:674–7. Fluids Eng, Trans ASME 1992;114(1):52–7.
[25] Syms GF. Simulation of simplified-frigate airwakes using a Lattice-Boltzmann [40] Okajima A. Strouhal numbers of rectangular cylinders. J Fluid Mech
method. J Wind Eng Ind Aerodynam 2008;96(6-7):1197–206. 1982;123:379–98.
[26] Spalart PR, Jou WH, Strelets M, Allmaras SR. Comments on the feasibility of LES [41] Norberg C. Flow around rectangular cylinders: pressure forces and wake
for wings, and on a hybrid RANS/LES approach. In: Advances in DNS/ frequencies. J Wind Eng Ind Aerodynam 1993;49:187–96.
LES. Greyden Press; 1997. p. 137–47. [42] Yu D-H, Kareem A. Numerical simulation of flow around rectangular prism. J
[27] Strelets M. Detached eddy simulation of massively separated flows. In: 39th Wind Eng Ind Aerodynam 1997;67 and 68:195–208.
AIAA aerospace sciences meeting and exhibit, Reno, NV; 2001. [43] Jeong J, Hussain F. On the identification of a vortex. J Fluid Mech
[28] Spalart PR, Squires KD. The status of detached-eddy simulation for bluff 1995;285:69–94.
bodies. In: Direct and large eddy simulation V. Springer; 2004. [44] Hicks BB. Propeller anemometers as sensors of atmospheric turbulence.
[29] Padfield GD, White MD. Flight simulation in academia: HELIFLIGHT in its first Bound–Lay Meteorol 1972;3:214–28.
year of operation at the University of Liverpool. Aeronaut J [45] Counihan J. Adiabatic atmospheric boundary layers: a review and analysis of
2003;107(1075):529–38. data from the period 1880–1972. Atmos Environ 1975;9(10):871–905.
[30] Forrest JS, Hodge SJ, Owen I, Padfield GD. Towards fully simulated ship– [46] Polsky SA. CFD prediction of airwake flowfields for ships experiencing beam
helicopter operating limits: the importance of ship airwake fidelity. In: winds. In: 21st AIAA applied aerodynamics conference, Orlando, FL; 2003.

View publication stats