lable at ScienceDirect

Renewable Energy 71 (2014) 742e754

Contents lists avai

Renewable Energy

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

Assessing the influence of inflow turbulence on noise andperformance of a tidal turbine using large eddy simulations

Thomas P. Lloyd a, b, *, Stephen R. Turnock b, Victor F. Humphrey b

a MARIN Academy, Maritime Research Institute Netherlands, Wageningen, The Netherlandsb Faculty of Engineering and the Environment, University of Southampton, Southampton, United Kingdom

a r t i c l e i n f o

Article history:Received 20 February 2014Accepted 7 June 2014Available online 17 July 2014

Keywords:Horizontal axis tidal turbineLarge eddy simulationInflow turbulence generatorAcousticsEnvironmental impact

* Corresponding author.E-mail address: T.P.Lloyd@soton.ac.uk (T.P. Lloyd).

http://dx.doi.org/10.1016/j.renene.2014.06.0110960-1481/© 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

Large eddy simulations of a model scale tidal turbine encountering inflow turbulence have been per-formed. This has allowed both unsteady blade loading and hydrodynamic noise radiation to be predicted.The study is motivated by the need to assess environmental impact of tidal devices, in terms of theiracoustic impact on marine species.

Inflow turbulence was accounted for using a synthetic turbulence generator, with statistics chosen torepresent the gross features of a typical tidal flow. The turbine is resolved in a fully unsteady mannerusing a sliding interface technique within the OpenFOAM® libraries. Acoustic radiation is estimated usinga compact source approximation of the Ffowcs WilliamseHawkings equation.

It is observed that the long streamwise length scale of the inflow turbulence results in characteristic‘humps’ in the turbine thrust and torque spectra. This effect is also evident in the far-field noise spectra.The acoustic sources on the blades are visualised in terms of sound pressure level and “Powell's sourceterm”. These measures show that the dominant sources are concentrated at the blade leading edgestowards the tip. This results from the high loading of the turbine blades, and causes the sound to radiatemore akin to a monopole than a dipole.

The full scale source level, obtained from scaling of the simulation results, is found to be lower thancomparable measured data reported in the literature; this is attributed to additional sources not includedin the present study. Based on the predicted source level, no physical impact on fish is expected.

© 2014 Elsevier Ltd. All rights reserved.

1. Introduction

The level of acoustic emission from tidal turbines is part ofenvironmental impact assessment [1]. It is possible that tidal tur-bine noise will have some impact on marine life, but attempts toquantify this are limited [2]. We direct the reader to work involvingfull-scale turbine noise measurement [3,4], as well as noise esti-mation for smaller devices [5,6]. Turbine noise sources arecommonly defined in terms of a sound pressure level (SPL)measured at some far-field distance. These are often correctedback to a source level at 1 m from the rotor. A typical leveldefined in this way is of the order of 166 dB re 1 mPa2 at 1 m [3].Richards et al. [4] expect the dominant noise from horizontalaxis tidal turbines (HATTs) to be due to rotating machinery in afrequency range z1e100 Hz.

Wang et al. [5] measured the noise of a 0.4 m diameter device,using a scaling procedure recommended by ITTC [7]. The reportedmaximum third-octave bandwidth SPLs (for a freestream velocityof 2.57 ms�1) were approximately 115 dB and 125 dB for model andscaled results respectively.

Numerical studies of tidal turbine noise are less commonly re-ported. The noise of a vertical axis tidal turbinewas estimated usinga discrete vortex method by Li and Çalisal [6]. These authors foundthe peak SPL occurs at 4 Hz, and related their findings to thehearing sensitivity of fish, without making direct environmentalimpact assessments. No studies of HATT noise have been located;by contrast, noise simulations of horizontal axis wind turbines aremore commonplace [8e10].

It is important to study the dynamic forces experienced by aturbine, since they contribute to fluid structure interaction effects,such as blade fatigue [11] or potential improvements in powercapture [12]. Thus studying this behaviour in a dynamic environ-ment would seem appropriate. The effect of inflow turbulence on

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754 743

turbine wakes will also influence the layout of arrays in order tooptimise total power capture [13,14].

The effect of inflow turbulence on the length of the turbinewakehas been studied using computational fluid dynamics [15,14].Various approaches to simulating unsteady loading have been used.Churchfield et al. [16] used a library of instantaneous turbulencerealisations as an inlet condition to replicate a turbulent boundarylayer. Alternatively, ‘synthetic’ turbulence can be generated using anumerical method to represent the inflow turbulence. In Ref. [15],this approach was used to show that the inflow turbulence reducesthe length of the turbine wake. This effect was also reported byMcNaughton et al. [14], who included the turbine geometry in thesimulation, as opposed to the porous disk representation in Gantand Stallard [15]. Afgan et al. [17] showed that large eddy simula-tion (LES) is more capable than an unsteady Reynolds-averagedNaviereStokes equations solution at predicting the complex flowfeatures and mean performance of a model scale turbine. LES alsoprovides better resolution of the inflow turbulence and thrustspectra.

We have previously presented predictions of tidal turbine noise[18] as well as impact assessment [19], using empirical modelling.The approach was based on modelling the turbine unsteady thrustspectrum due to inflow turbulence, and predicting the noiseassuming free field radiation [20]. Inflow turbulence is known to bethe dominant noise source due to turbulence in this case [18,21].The estimated spectral source level (SSL) was z145 dB re1mPa2 Hz�1 at 1 m across a frequency range of z10e100 Hz. Thispaper presents a development in terms of noise simulation ofHATTs, and is based on the methodology in Lloyd et al. [22].

The paper assumes the following format. x2 outlines the nu-merical setup for the simulations, including the methods forgenerating inflow turbulence and predicting sound radiation. In x3,the test case geometry is described, along with the domain design.A model scale turbine is used since a detailed blade geometry isavailable, as well as mean performance data. This section also in-cludes grid design considerations and presents an assessment ofthe sliding interface technique used in terms of its ability tointerpolate the broadband velocity fluctuations present in theinflow. The simulation results are divided into three sections:inflow turbulence statistics (x5); turbine response (x6); andacoustic emission (x7). The model scale acoustic predictions arecompared to an analytical model, since no experimental validationdata is available. These are then scaled using recommended pro-cedures in order to provide estimates of full scale turbine noise (x8).This allows discussion of possible environmental impact. Finally,conclusions are made in x9.

1 www.openfoam.org/.

2. Numerical framework

2.1. Turbulence modelling

In large eddy simulation, the filtered NaviereStokes equationsare resolved in a time-dependent manner. Scales smaller than thegrid are accounted for using a subgrid model. For a detaileddescription of large eddy simulation see for example Sagaut [23].The dynamic mixed Smagorinsky subgrid model [24] is used, sinceit has been shown to performwell for complex flows and on coarsegrids [25].

The normalised first cell height is defined as Dyþw ¼ y1ut=n,where y1 is the first cell height, ut the friction velocity and n thekinematic viscosity. An average value of Dyþw ¼ 40 was achievedover the blades; due to the surface refinement technique employedby snappyHexMesh, the leading and trailing edges typicallypossessedmuch lower values. Since the viscous sublayer is not fully

resolved (this requires Dyþwz1), awall function, based on Spalding's“law of the wall” [26] was used.

In order to simulate stochastic loading on the turbine, an inflowturbulence generator is used. This is a numerical method forgenerating synthetic turbulence at the simulation inlet. Here weuse the forward stepwise method; for a full description of themethod, see Kim et al. [27]. Evaluations of this method for hydro-acoustic predictions has previously been made.

2.2. Solution method

Simulations were performed using the OpenFOAM®1 libraries. Acustom solver based on the pimpleDyMFoam application was used.The main features of the solver are: pressure implicit splitting ofoperators (PISO)-type [29] correction of the velocity; outercorrector loops allowing higher time steps than PISO; grid rotationvia ‘dynamic meshing’ and an arbitrary mesh interface (AMI); andvelocity fluctuations generated by the FSM inserted during the PISOloop. All discretisation schemes are second-order, apart fromconvective acceleration, which uses a hybrid upwind-central dif-ferencing scheme, giving good accuracy in regions where a centralscheme is less accurate [25,28]. Linear solution was achieved usingthe biconjugate gradient method for velocity, and general algebraicmultigrid method for pressure. The solvers exit the iteration loopwhen a tolerance of 10�9 (velocity) and 10�6 (pressure) is achievedwithin each loop.

The pimpleDyMFoam solver allows the maximum Courantnumber Co ¼ jujDt=Dx to exceed unity, where juj is a local velocitymagnitude, Dt the time step and Dx the cell dimension; simulationsused a maximum time step Dt� ¼ DtU0=D ¼ 3:5� 10�5, based onthe reference (freestream) velocity U0, and turbine diameter D.where the maximum Courant number was also limited to four. Thistime step is the same as that used for other tidal turbine LES [17],and results in 40 time steps per degree of rotation. A transientphase of four turbine rotations was assumed (T� ¼ TU0=Dz2:3),allowing the inflow turbulence to reach the rotor plane. Probe, forceand sound pressure were then sampled at fsample ¼ n/300, or 100times per blade passage, for a further T�z6:9, thus ensuring acomplete flow-through of the domain.

2.3. Acoustic analogy

In order to evaluate the acoustic radiation from the blades, weuse a formulation of the FfowcsWilliam-Hawkings (FW-H) acousticanalogy [30]. This has been implemented into OpenFOAM® usingonly the term relating to fluid loading, which is an acoustic dipole.This is given by

p'ðx; tÞz xi

4pc0���r���2

v

vt%SnjpijdðhÞdSðyÞ: (1)

In Equation (1), p0 is acoustic pressure, x and y denote thereceiver and source locations, jrj ¼ jx� yj, c0 is the speed of sound,equal to 1500 ms�1 inwater, nj is the normal vector to the surface S,and d is the Dirac delta function. This form of the FW-H equation issuitable for low Mach number flows where broadband noise is ofinterest. It assumes the receiver to be in the acoustic far-field(jrj[l) and the source to be compact (L≪l, where L is the sourcedimension). The integration surface (corresponding to h ¼ 0) istaken to be the solid boundary.

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754744

A receiver distance of jrj ¼ 2D was used, since it lies in theacoustic far-field for a typical rotor [31], but still within the rangethat environmental impact is possible [18]. Sound pressure levelpredictions were made at three receiver angles of q ¼ 0�, 45� and90�, where 0� corresponds to the rotor axis, downstream of therotor plane. The SPL is defined as

SPLðf Þ ¼ 10log10

0@p02ðf Þ

p20

1A; (2)

where p0 ¼ 1 mPa in water. The pressure fluctuation p02ðf Þ wasestimated using Welch's algorithm [32], applied to the time tracesobtained from Equation (1).

3. Test case description

3.1. Turbine geometry

The turbine geometry used is a representation of the threebladed model scale rotor tested by Bahaj et al. [33]. The blades useNACA 63�8xx sections with a reduction in chord and thicknessfrom root to tip. Parameters for the chosen case are given in Table 1.The chosen case represents a high thrust loading. Tip speed ratio isdefined as L ¼ UR=U0 where U is the rotational velocity in radiansper second, R is the turbine tip radius and U0 is the freestreamvelocity.

3.2. Specifying inflow turbulence statistics

Characterisation of the inflow turbulence was achieved byaiming to replicate features of full scale environmental turbulence.The chosen statistics are based on IEC standard 61400�1, and arethe same as those used by Gant and Stallard [15]. The horizontalintegral length scale is then L x;z ¼ 0:7D, with L y ¼ L x;z=6. Aturbulence intensity of I ¼ 10% is used, which is specified as ho-mogeneous and isotropic. Note the FSM has the ability to generatefully inhomogeneous, anisotropic mean velocity, length scale andturbulence intensity profiles more akin to realistic tidal flows.However, as a demonstration of our methodology, only anisotropiclength scales are used here.

4. Grid design

4.1. Domain setup

Since the data presented by Bahaj et al. [33] are corrected forblockage effects, the turbine is simulated in an open domain. Theblockage ratio (ratio of turbine rotor disc area to domain cross-section area) is 0.022. Based on corrections made for a similarblockage ratio, reported by Walker et al. [34], the effect on turbineperformance is expected to be small. The grid consists of two re-gions, the rotor and stator, with the rotor region fully encompassing

Table 1Summary of key tidal turbine test case parameters.

Symbol Meaning Value Unit

D Rotor diameter 0.8 mB Number of blades 3 e

U0 Mean freestream velocity 1.4 ms�1

n Rotational velocity 3.29 s�1

L Tip speed ratio 5.96 e

gr Blade root twist angle 15 deggT Blade tip twist angle 0 deg

the turbine geometry. Geometrical simplifications of theturbine geometry result in a truncated hub of length and diameterdH/D ¼ 0.125. The rotor grid has dimensions of LI/D ¼ 0.625 anddI/D ¼ 1.2, and rotates inside the stator grid at a constant rotationalvelocity. The two grid regions are connected using an arbitrarymesh interface (AMI), which interpolates variables between thepatch faces of each grid region using Galerkin projection. Fig. 1shows a schematic of the domain layout, with associated bound-ary conditions given in Table 2.

The background cell size is chosen to resolve z80% of the totalturbulence kinetic energy, with a grid cutoff size of DzL =12 [35].This ensures that the cutoff lies inside the inertial subrange of theturbulence spectrum, with an estimated cutoff frequency ofz47 Hz. The maximum resolvable acoustic frequency is expectedto be higher however, due to the convection velocity through therotor [36]. Based on L ¼ 5:96, an acoustic spectrum up toz280 Hzis achievable.

The grid was created using the blockMesh and snappyHexMeshutilities within the OpenFOAM® libraries. The resulting unstruc-tured grid has a total of z4.6 M cells, consisting of approximately90% hexahedral volumes, with the remaining cells being polyhedra.Views of the grid are provided in Fig. 2.

4.2. Sliding interface resolution

It is important to consider the effect of the rotating grid on theresolved turbulence. If eddies do not convect through the interfaceaccurately, the rotor will not see the correct turbulence spectra.Fig. 3(a) shows turbulence structures convecting through the AMIwith only limited numerical dissipation, which is consistent withthe findings of Bensow [37]. The largest interpolation error is seenat the outer limits of the AMI, which is not expected to have alarge effect on the fluctuations experienced by the rotor blades.Qualitative comparison of the streamwise slices on either side ofthe AMI (shown in Fig. 3(b) and (c)) shows small scale turbulencepassing into the rotor region.

Streamwise velocity spectra in Fig. 4 show that the range ofresolved scales does not appear to be affected by the AMI. The gridcutoff of 2D has been realised, corresponding to the desiredfrequency of z47 Hz. An additional measure of the interpolationquality at the AMI is a Courant number based on the turbinerotational velocity, i.e.

Co ¼ URIDtDx

; (3)

where RI is the radius of the AMI. Based on a cell size Dx ¼ 0.015 m,this results in Co z 0.0069. This is approximately 11 times smallerthan the resolution found to be sufficient for laminar vortexshedding of a circular cylinder [38], and similar to that used byAfgan et al. [17] for the same tidal turbine. Thus the resolution usedin the present study is deemed to be suitable for assessingbroadband noise.

5. Inflow turbulence statistics

Table 3 summarises turbulence statistics sampled upstream ofthe turbine rotor plane. Turbulence intensity is higher than thedesired value of 10% at both centreline probe locations. This ispartially related to the increased value of I specified at the inflowplane in order to account for streamwise decay. However, theincrease in Ix from probe locations 1 to 2 may be attributed to theblockage caused by the turbine. The value of L x at probe location 1is also a result of the inflow generator, which has been shown togive larger integral length scales than specified [28]. A reduction in

Fig. 1. Schematic showing domain layout (not to scale). Patch names relate to boundary conditions in Table 2. Numbered filled circles denote probe locations one to four. Globalorigin at turbine rotor plane, on rotor axis.

Table 2Tidal turbine boundary conditions. Patch locations relative to global origin centredon turbine rotor plane.

Patch Location Velocity Pressure

Inlet �3D Fixed value Zero gradientOutlet 7D Convective Fixed value zeroSides 3D Symmetry SymmetryTop 3D Symmetry SymmetryBottom 3D Symmetry SymmetryBlades Moving wall Zero gradientHub Slip wall Zero gradient

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754 745

length scale is seen at probe location 2, also due to the turbineblockage effect.

A visualisation of the flow is provided in Fig. 5, showing tur-bulence structures via two measures: u�

x ¼ uxD=U0 is the normal-ised streamwise vorticity, plotted on a vertical slice; Q is the secondinvariant of the velocity gradient tensor, which provides a scalarfield for vortex identification. The snapshot has been taken atT* ¼ 1.3, which corresponds to the transient phase of the simula-tion, in order to show the distinction between the inflow and waketurbulence. On the left of the figure, the large streamwise lengthscale of the inflow turbulence may be observed; based on thespecified parameters, scales as large as the rotor diameter shouldexist. Behind the rotor plane, tip vortices are seen to convectthrough the AMI. Vortices also form from the turbine hub. In thenear wake region (x/D < 5) the wake structure appears fairlycoherent, while further downstream, the increase in turbulencemixing leads to more fine scale structures.

Fig. 6 illustrates this further, using a perspective view and twodifferent isosurface values for Q: the value Q ¼ 0.5 ms�1 allows thelarge inflow turbulence structures to be seen; while the Q¼ 5 ms�1

contour reveals the turbine wake vortices.

Fig. 2. Views of tida

6. Unsteady loading

Performance assessment of the turbine is made in terms ofthrust and power. Fig. 7 depicts time traces of the turbine thrustand power, while Fig. 8 shows the associated spectra. The instan-taneous thrust and power coefficients monitored during thesimulation were calculated as

CT ðtÞ ¼2FxðtÞr0AU2

0

(4a)

and

CPðtÞ ¼2UMxðtÞr0AU3

0

; (4b)

where Fx is the thrust (streamwise force), and Mx is torque. Timeand frequency have been normalised using the rotation rate n. Themean values for both numerical and experimental results areincluded in Fig. 7, as well as in Table 4. Also included are results forthe same case obtained using blade element momentum theory(BEMT) by Banks et al. [39], and using LES [17]. An increase in thrustcoefficient between the LES cases with and without inflow turbu-lence is seen. No corresponding increase in power is observedhowever. This agrees with similar studies from the literature[14,17]. In comparison to the BEMT results, the LES value of CT iscloser to the experiment, while for CP the BEMT shows betteragreement. This is due to the use of wall functions in the LES, which(in OpenFOAM®) have been shown to under-predict drag comparedto a wall-resolved boundary layer [40]. Since the BEMT includesempirical blade section lift and drag data, this effect is not asprominent when using this method. Afgan et al. [17] found that agrid of 5�106 cells under-predicted thrust by z6% compared to a

l turbine grid.

Fig. 3. Effect of AMI on resolved turbulence: normalised streamwise velocity u� ¼ u=U0, upstream of turbine.

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754746

21 �106 cell grid. This may be a further reason for the discrepancyin the mean thrust prediction.

Root mean square performance coefficients from the presentstudy are compared to those presented by Afgan et al. [17]. Theyreported results for I ¼ 1, 10 and 20%, at g ¼ 20� and L ¼ 6. Thedata presented in Table 4 are for I z 20%, since this is closest to thepresent study. The values for the present study are higher than inRef. [17], who simulated 24 turbine rotations, compared to 16performed here. This suggests that a longer statistical sampling

Fig. 4. Streamwise velocity spectra at probe locations upstream and downstream ofarbitrary mesh interface: probes 1 and 2 located at x=D ¼ �0:25 and �0:375 on thedomain centreline.

period would reduce rms coefficients. The magnitudes of the rmsvalues for the LES casewith inflow turbulence are approximately 30times higher than for the case without. Afgan et al. [17] presentedsimilarly small values for a case with I z 1%.

Two main features of the time traces of thrust and power areevident in Fig. 7. The slowly varying part is associated with thepassage of the largest length scales; these have a period of just over1 s, corresponding to a length scale approximately twice the inte-gral length. The higher frequency fluctuations may be attributed tothe blades ‘cutting’ through long streamwise eddies. This results inblade-to-blade correlation of the thrust and torque.

The spectra presented in Fig. 8 are characterised by ‘humps’known as haystacks, close to the blade passing frequency (BPF; 0thharmonic) and first harmonic (at approximately 10 and 20 Hzrespectively). The decibel difference between the numerical spectraand representative smooth curve at these frequencies is indicated

Table 3Summary of velocity probe data for tidal turbine in open domain. Probe locationsshown in Fig. 1.

Probe 1 2 3 4

x=D �0.375 �0.25 �0.25 0.25y=D 0.0 0.0 0.35 0.35u 1.20 0.97 1.11 0.81Ix=% 13.8 15.7 19.3 24.0Iy=% 12.7 15.2 17.0 21.0Iz=% 15.0 20.2 19.7 21.6Lx=m 0.81 0.43 0.3 0.21

Fig. 5. Flow visualisation of tidal turbine in open domain at T� ¼ 1:3: isosurface of Q ¼ 10 s�1 and x-y plane slice of u�x at domain centreline. Inlet on left.

Fig. 6. Flow visualisation of tidal turbine in open domain at T� ¼ 1:3: isosurfaces of Q ¼ 0:5 s�1 (inflow turbulence) and Q ¼ 10 s�1 (turbine wake) and x-y plane slice of u� atz=D ¼ 2. Inlet on left.

Fig. 7. Time trace of thrust and power coefficient for cases with and without inflowturbulence. Time non-dimensionalised as t� ¼ tn. Dashed and dotted lines show nu-merical and experimental mean values, with averaging duration indicated by length ofline.

Fig. 8. Power spectral density of time traces. Frequency normalised using the bladepassing frequency (Bn). Dashed line indicates equivalent smooth spectrum (L x≪P);dotted line denotes blade passing frequency and 1st harmonic.

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754 747

Table 4Comparison of turbine performance coefficients: mean (top) and rms (bottom).Values from BEMT reported in Banks et al. [39]. Large eddy simulation cases with (y)and without (z) inflow turbulence.

Case Exp. LESy LESz BEMT

CT 1.00 0.98 0.90 0.86CP 0.36 0.43 0.43 0.37

Case LESy LES [17] LESz

C0T ;rms 0.0535 0.0039 0.0015

C0P;rms 0.0510 0.0024 0.0017

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754748

in the figure. A magnitude of 10log10(B) (or z4.8 dB) is also pre-dicted by Blake [20, chap. 10] using Equations A.1a and A.1b. This isdue to the shape of the admittance function (Equation (A.5)), whichapplies only when L x[P=B.

An estimate of the turbine hydrodynamic pitch can be madeusing the local resultant flow velocity seen by a blade section. Aradius r ¼ 0.7R has been used, with streamwise and tangentialinflow factors of ax ¼ 0.32 and aq ¼ 0.025 estimated using BEMT(taken from Ref. [13]). The pitch is given by P/D ¼ ptan(4), where4¼ aþ g is the hydrodynamic pitch angle and a is the local angle ofattack. This results in P/B z 0.134 m, which confirms that themagnitude of the haystacks seen in Fig. 8 is due to the streamwiseintegral length scale exceeding the rotor pitch. Jiang et al. [41] useda similar analytical model to analyse propeller broadband forces.They emphasised the presence of only two prominent haystacks,whose peaks are skewed to slightly higher frequencies than the BPFand first harmonic. This effect can be seen in Fig. 8.

The haystacking observed in the thrust and power spectra iselucidated by comparing total rotor and single blade thrust, pre-sented in Fig. 9. The combined thrust from all three blades is alsoshown; this is equivalent to increasing the single blade thrust by10log10(3). Note that the dashed line is comparable to the dashedline in Fig. 8, which corresponds to a response spectrum whereL x < P. Summing the blade thrust spectra removes phase infor-mation relating to blade-to-blade correlation, and hence thespectral humps are not captured.

7. Model scale noise predictions

The acoustic sources on the blades are viewed as the soundpressure level on the wall (SPLw), which is defined as

SPLw ¼ 10log10

0@p02wp20

1A: (5)

Fig. 9. Comparison of thrust spectra for single blade, all blades and rotor. Spectral levelfor all blades is 10logð3Þ higher than for single blade. Dashed line equivalent to dashedline in Fig. 8.

Contours of SPLw on the suction sides of the blades for caseswithandwithout inflow turbulence are given in Fig.10. Avalue range hasbeen chosen to highlight the difference between the two cases;hence some of the detail of the source distribution in Fig.10(a) is notshown. Fig. 10(a) shows a higher source level across most of theblade span, with the highest source amplitude located on the outerpart of the blade. The exact spanwise and chordwise location of thehighest SPLw is unclear however, due to much of the source levelexceeding themaximumvalue of 130 dB. This is addressed in Fig.11.

In order to locate the dominant acoustic source more accurately,the SPLw has been re-scaled and displayed close to the tip(r/R ¼ 0.8�1.0). Fig. 11 reveals that the acoustic source is centred inthe outer region of the blade, but not at the tip (r/R z 0.9). Thiscorresponds to the location of peak spanwise loading for a typicalturbine blade [13], and is similar to that expected for blade trailingedge noise [42]. In addition, the source is concentrated at the bladeleading edge. This was expected based on the source distributionsexhibited for typical wind turbine aerofoils [43]. Certain locationsin Fig. 11(a) do show a higher source level towards the trailing edgehowever. This is also revealed in Fig. 12, and may be explained byflow separation. A final point to note is the higher SPLw on thesuction side of the blade, which indicates that the far-field noisemay be higher downstream of the device.

Acoustic sources can also be visualised in terms of vorticity. Thisfollows from the fact that surface pressure fluctuations result fromthe passage of eddies close to the surface. Here we use “Powell'ssource term” [44], which is a volume dipole source defined as

P ¼ V,ðu� uÞ; (6)

where u is the vorticity vector. This measure has been used pre-viously to examine the effect of wind turbine blade tip shape onnoise [8]. The difference between the source distributions is clearwhen comparing Fig. 12(a) and (c). A localised region of the iso-surface towards the blade trailing edge (see Fig. 12(a) insert) isassociated with blade separation, which results in an additionalnoise source. The separation is caused by a region of high velocityturbulence impinging onto blade 1, as shown in Fig. 12(d). Thisshifts the local relative velocity, increasing the angle of attack,leading to flow separation. When the blade is not experiencing anincreased inflow velocity, the size of this separated region isreduced, as shown in Fig. 12(b).

It is known that the highest overall SPL for inflow turbulencenoise occurs at q ¼ 0�, based on the dipole assumption; this data ispresented in Fig. 13, with comparison to the analytical model ofBlake [20], chap. 10, the formulation of which is included inAppendix A. The numerical result compares favourably with theanalytical model. The maximum discrepancy between numericaland analytical SPL is 5 dB, which occurs at the BPF. Differencesbetween the two analytical scenarios presented are also clear,especially at low frequencies. Haystacks at the BPF and associatedharmonics are well captured by the simulation. The intermediatehumps are however not as clearly visible in the numerical spectrumas the analytical; this is partly related to the bandwidth of the fastFourier transform and is most evident at low frequencies. This partof the spectrum would be more accurately predicted if the totalsimulation duration were increased.

Noise directivity is examined in Fig. 14. Fig. 14(a) compares thesimulation result presented in Fig.13 to predictions made at q¼ 45�

and 90�. It is evident that the expected behaviour of a pure acousticdipole, assumed in the analytical model, is not fully realised in thesimulation. At the BPF, a difference of approximately 20 dB is seenbetween receiver angles of 0� and 90�. This is also shown in termsof the overall sound pressure level (OASPL), plotted in Fig. 14(b).The OASPL is defined as

Fig. 10. Acoustic source on suction (downstream) side turbine blades for open domain, visualised as surface sound pressure level SPLw ðp0 ¼ 1m PaÞ.

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754 749

"Z f2p02df

#

OASPL ¼ 10log10

f1

p20; (7)

which is the decibel level of the normalised acoustic energy acrossthe frequency range f1 � f2. The reduction in OASPL between 0� and90� is z16 dB.

Morton et al. [45] attributed this monopole-like behaviour toincreased tip loading at low advance coefficients. The turbineadvance coefficient (J ¼ p=L ¼ 0:53) is slightly lower than thatused byMorton et al. [45], where J¼ 0.7. This may explain the largerincrease in level observed, compared toMorton et al. [45], who sawa 10 dB difference between the same angles. Hence there is po-tential for further investigation of the effect of turbine operatingcondition on noise directivity. Despite this, Fig. 14(a) provides evi-dence of the improved predictive capabilities of the simulationcompared to the analytical model used here.

Cases with and without inflow turbulence are now compared.The noise directivity without inflow turbulence is shown in Fig. 15,

Fig. 11. Acoustic source on turbine blade tip, visualised as surface sound pressure levelSPLw ðp0 ¼ 1m PaÞ.

along with the data for the case with inflow turbulence at q ¼ 0�

(already presented in Fig. 13). Since a 40 dB difference in SPL existsat all frequencies between the two data sets at q ¼ 0�, it may beconcluded that the loading noise in the turbine axis is negligiblewhen the inflow is steady. This would be expected, since there areno incoming velocity fluctuations to generate unsteady thrust onthe rotor. However, at q ¼ 45 and 90�, there is clear evidence of atonal noise component. This dominates at the BPF, but also appearsto have some frequency content at f ¼ n, and is known as steadyloading, which is known to be negligible for subsonic rotors [[46],chap. 3]. This is shown by the z20 dB difference between themaximum steady loading noise (at q ¼ 90�) and the maximumunsteady loading noise (at q ¼ 0�).

8. Full scale noise

8.1. Acoustic scaling

In order to carry out environmental impact studies, full scaleturbine source levels are required. This data may be used toinvestigate animal response experimentally [1] or make noiseimpact predictions at the turbine design stage [18]. Full scale tur-bine noise has been assessed using two approaches: scaling of themodel scale simulation results; and use of Blake's model for a fullscale turbine, as previously presented in Lloyd et al. [18]. Thescaling procedure was first presented in Ref. [22], and is similar tothat recommended by ITTC [7], and by Wang et al. [5] for tidalturbine measured noise.

Rudimentary scaling is based on the Strouhal number

St ¼ fDffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiU20 þ U2R2

q (8)

and the acoustic intensity

I ¼ p02

r0c0: (9)

Denotingmodel and full scale values by the subscripts 'M' and 'F',the frequency scales as

fF ¼ fMnFnM

: (10)

The acoustic intensity can be assumed to scale as

I∝r0u5ℒ

2

c20jrj2; (11)

Fig. 12. Distribution of Powell's sound source close to turbine blades: isosurfaces of P ¼ V,ðu� uÞ ¼ 5� 106 s�1 (black) and rotor plane slices of normalised instantaneous streamwise velocity u�.

T.P.Lloydet

al./Renew

ableEnergy

71(2014)

742e754

750

Fig. 13. Sound pressure level for model scale turbine in open domain at q ¼ 0+ andjrj ¼ 2D, predicted using FW-H equation and Blake's analytical model. Df ¼ 1 Hz.Dashed line indicates equivalent smooth analytical spectrum ðL x≪PÞ; dotted linedenotes analytical spectrum where L x[P=B.

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754 751

following Howe [[47], chap. 3]. Hence, taking u ¼ UT (the tipvelocity),

p02F ¼ p02MUF

UM

� �5 ℒF

ℒM

� �2 jrjMjrjF

� �2: (12)

It has been assumed that the speed of sound is constant be-tween model and full scale. The result of applying this scalingprocedure to the model scale data is shown in Fig. 16. Blake's modelusing full scale parameters is also included. The full scale turbine isassumed geometrically similar to the model scale device, with arotor diameter of 22m; this is reasonable for installed turbines [48],and has been used for both empirical [19] and numerical [22]studies. The tidal velocity is taken to be 2.5 ms�1, with the sameturbulence characteristics as described in Section 2.

Fig. 16 is evidence that the scaling procedure is reasonably ac-curate for this simple case. The cutoff frequency, as a result of thescaling procedure, reduces toz14 Hz. In order to allow comparisonbetween model and full scale data, and to published source levels

Fig. 14. Noise directivity for mod

(which may use a different bandwidth), indicative full scale soundlevels are provided in Table 5.

The 1 Hz bandwidth source level of 1441 mPa2 m2 Hz�1 isz6 dBlower than the peak value predicted from measurement data byWang et al. [5], using a similar scaling method. This discrepancy islikely due to cavitation noise, which was observed in the experi-ments, and has not been simulated. Suction side leading edgecavitationwas seen byMolland et al. [49], who tested the NACA 63-815 section used to design the turbine blade simulated here.

A further noise source is expected to bemechanical noise, whichresults from the drive train components of the turbine. Account forthis noise source can be made using an empirical model by Lloydet al. [19]. In this case, the resulting third-octave turbine soundlevel isz160 dB re 1 mPa2 Hz�1 at 1m. In Lloyd et al. [19], the sourcelevel of the mechanical and hydrodynamic noise sources wasestimated to be approximately equal. Assuming the two sources areincoherent, their sound pressure levels may be combined as

SPLAB ¼ 10log10

0@10

SPLA10 þ 10

SPLB10

1A; (13)

where ‘A’ and ‘B’ are two different incoherent sources. This resultsin a z3 dB increase in combined source level.

8.2. Environmental impact and mitigation

The acoustic predictions presented in Fig. 16 may be used toperform environmental impact assessments. Here a simplifiedassessment is made based on the source level presented in Table 5.The species most likely to be affected are fish, due to their hearingsensitivity being higher at low frequencies [1]. Li and Çalisal [6]found that vertical axis tidal turbine noise peaked at 4 Hz, butdid not present an impact assessment. The peak of inflow turbu-lence noise simulated here occurs at <1 Hz. Since this is below thelowest frequency available from hearing threshold data for marinespecies [4], it is hard to assess its environmental impact. The cutofffrequency of the spectrum does however lie within the range ofhearing threshold data. At 10 Hz, typical values for fish’ hearingthreshold and ocean background noise are 80 dB re 1 mPa2 [4] and

el scale turbine at jrj ¼ 2D.

Fig. 15. Noise directivity for model scale turbine: case without inflow turbulence, atjrj ¼ 2D. Steady loading (Gutin sound) indicated at rotation rate (n) and blade passingfrequency (BPF). Data for case with inflow turbulence at q ¼ 0+ included for compar-ison (turb).

Table 5Full scale turbine noise levels at blade passing frequency, chosen to be representa-tive of peak spectral level. Sound pressure level from scaled data.

Quantity Df Correction Value Unit

SPL 0.01 e 88 dB re 1mPa2Hz�1

SPL 1 e 108SSL 0.01 20log10ðjrj � 1Þ 124 dB re 1m Pa2 Hz�1

at 1 mSSL 1 10log10ðDf Þ 144third octave SSL 0.4 10log10ðDf Þ 140

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754752

75 dB re 1 mPa2 Hz�1 [[50], chap. 7] respectively. The latter value istypical of shipping noise. Although the simulation does not resolvespectra up to the highest frequency of interest in terms of envi-ronmental impact (z100 Hz), the maximum amplitude of the hy-drodynamic noise is captured.

Based on the SPL of 108 dB re 1 mPa2 Hz�1 quoted in Table 5, nohearing threshold shift would be expected. This requires the SPL toexceed the species' hearing threshold by at least 75 dB for 8 hourwithin a 24 hour period [4]. Hearing threshold shift has been pre-dicted by Ref. [19] however; accounting for mechanical noise aswell as an array of three turbines, a third-octave SPL of 140 dB re1 mPa2 Hz�1 was estimated at a frequency of 160 Hz. This agreeswith full scale measurements of tidal turbine source levels used tocarry out environmental impact assessments [1,3], and suggeststhat mechanical noise may be more important at higher fre-quencies than inflow turbulence noise.

Since the unsteady inflow conditions experienced by installedturbines are difficult to avoid, noise reduction of devices wouldideally focus on reducing tip speed, due to the fifth power de-pendency of acoustic pressure on velocity. However, as turbinestypically have an optimum tip speed ratio of 5e6 [33], this strategywould lead to a reduction in turbine efficiency. Thus we recom-mend to reduce the turbine diameter, while maintaining a constanttip speed ratio. Due to the fact that turbine power is proportional toD2u3, the energy generation of smaller turbines will reduce

Fig. 16. Sound pressure level for full scale turbine in open domain at q ¼ 0+ andjrj ¼ 2D, predicted using scaling method and Blake's analytical model. Df ¼ 0:01 Hz.For SPL comparable to model scale data presented in Fig. 13, see Table 5.

significantly. This may be counteracted by the installation of mul-tiple devices [51]. A move towards utilising numerous smallerturbines would reduce overall noise radiation per unit of powergenerated. Assuming that all turbines have the same source level,the noise due to an array of turbines would only increase by10log10(ND) compared to a single device, where ND is number ofdevices. Large scale arrays may cause masking however, affectinganimal communications [2].

9. Conclusions

A simulation approach for predicting tidal turbine hydrody-namic noise has been developed and evaluated. The simulationsutilise three key components not used together before in the cur-rent application. These are: numerically generated inflow turbu-lence; fully resolved turbine geometry using large eddy simulationand an arbitrary mesh interface; and acoustic predictions via theFfowcs WilliamseHawkings equation.

The spatial and temporal quality of the interpolation at theinterface was assessed and found to be sufficient to allow the un-steady inflow to convect onto the rotor. The inflow turbulencecharacteristics used (anisotropic length scales and isotropic tur-bulence intensity) captured the gross features of the turbineresponse. However, the simulations could be developed to includeinhomogeneous turbulence statistics more similar to a tidalchannel.

Acoustic predictions were shown to be in good agreement withan analytical model, and exhibited characteristic haystacks causedby blade-to-blade correlation of the thrust response. The dominantacoustic sources have been shown to be concentrated at towardsthe blade tips, due to the high loading condition of the turbine. Thisalso causes the noise to radiate more akin to a monopole than adipole source. Estimates of full scale turbine noise were derivedusing rudimentary scaling procedures, which were shown to agreewith analytical estimates. The derived source level of 144 dB re1 mPa2 Hz�1 at 1 m is not expected to cause physical impact to fish.

The reported simulations may also be used to assess dynamicloads for blade design, such as root bending moment, although thiswas not the aim of the present study. It is expected that this type ofsimulation may be extended to allow fluid structure interactionanalyses in the future.

Acknowledgements

The authors wish to thank Dr. Yusik Kim for providing the inflowturbulence generator code. Computations were performed usingthe IRIDIS 4 cluster at the University of Southampton. The financialsupport of dstl, QinetiQ and the University of Southampton isgratefully acknowledged.

Nomenclature

LatinA rotor disc area [m2]B number of blades [�]

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754 753

Co Courant number [�]CP power coefficient [�]CT thrust coefficient [�]c0 sound speed [ms�1]D rotor diameter [m]f frequency [s�1]I acoustic intensity [kgs�3]I turbulence intensity [%]J advance ratio [�]L length [m]L integral length scale [m]n rotation rate [s�1]P blade pitch [m]P Powell's source term [s�2]p pressure [kg m�1s�2]p0 reference pressure [kg m�1 s�2]Q second-invariant of velocity gradient tensor [s�1]R turbine tip radius [m]r radius [m]jrj source-receiver distance [m]T time [s]Dt time step [s]U0 reference velocity [ms�1]u (streamwise) velocity [ms�1]x streamwise position [m]Dx cell dimension [m]Dyþw non-dimensional first grid cell height [�]

Greekg blade twist angle [deg]q receiver angle [deg]L tip speed ratio [�]l wavelength [m]r0 fluid density [kg m�3]U angular velocity [rad s�1]u angular frequency [rad s�1]u vorticity vector [s�1]

Subscripts0 reference valueF full scaleI interface valueM model scalerms root mean squarew wall value

Superscripts* normalised0 fluctuation- mean

AcronymsAMI arbitrary mesh interfaceBEMT blade element momentum theoryBPF blade passing frequencyHATT horizontal axis tidal turbineITTC international towing tank conferenceLES large eddy simulationOASPL overall sound pressure levelPISO pressure implicit splitting of operatorsSPL sound pressure levelSSL spectral source level

Appendix A. Analytical thrust loading model

As no experimental noise data is available for comparison, ananalytical model for thrust loading noise has been used [20, chap.10]. It assumes one of two forms, depending on the ratio of theintegral length scale to the rotor pitch. It is also a free-field model,i.e the effects of solid boundaries on the flow and acoustic radiationare not taken into account. The mean square thrust for a bandwidthDf is approximated as

T2ðf ;Df Þ ¼8<:16p3L �

R3R

Dfn

�qcJ

�2u02

U20

���S�u�cR

����2F�u�L �

q

�

����Aðu�Þ

���2;L x > P (A.1a)

T2ðf ;Df Þ ¼ 16p3BL �R

3RDfn

�qcJ

�2u02

U20

���S�u�cR

����2F�u�L �

q

�;L x≪P

�B

(A.1b)

where

u� ¼ u�U; L �

R ¼ L R�R and L �

q ¼ L q

�R (A.2)

are the normalised angular frequency, and normalised radial andcircumferential integral length scales;

J ¼ U0�nD; n ¼ U

�2p and q ¼ 1

�2r0U

20 (A.3)

are the advance coefficient, the rotational frequency and the dy-namic pressure; c and u02 are the blade tip chord and mean squarevelocity fluctuation; and P and B are the pitch and number of bladesrespectively. The Sears and admittance functions are given by

����S�u�cR

�����2z 11þ puc=UR

(A.4)

and

����Aðu�Þ����2 ¼

sinðpu�ÞexpfiðB� 1Þgpu�B

sin�pu�B

� : (A.5)

The length scale function is

F u�L �q

� � ¼ L �q

1þ u�L �q

� �2 : (A.6)

Integral length scales in Equation A.1a and A.1b are approxi-mated as a circumferential average of the Cartesian length scalesL 0 used in the simulations, giving L R ¼ L qz0:35 m.

The far field acoustic mean square pressure is derived from T2

using

p02 jrj; f ;Dfð Þ ¼ kcosq4pjrj

� �2T2 f ;Dfð Þ; (A.7)

where jrj is the receiver distance and k ¼ 2p/l the wavenumber.

T.P. Lloyd et al. / Renewable Energy 71 (2014) 742e754754

References

[1] Halvorsen M, Carlson T, Copping A. Effects of tidal turbine noise on fishhearing and tissues. Tech. Rep. PNNL-20786. Sequim: Pacific Northwest Na-tional Laboratory; 2011.

[2] Frid C, Andonegi E, Depestele J, Judd A, Rihan D, Rogers SI, et al. The envi-ronmental interactions of tidal and wave energy generation devices. EnvironImpact Assess Rev 2012;32(1):133e9. http://dx.doi.org/10.1016/j.eiar.2011.06.002.

[3] Parvin S, Workman R, Bourke P, Nedwell J. Assessment of tidal current turbinenoise at the Lynmouth site and predicted impact of underwater noise atStrangford Lough. Tech. Rep. Subacoustech Ltd.; 2005

[4] Richards S, Harland E, Jones S. Underwater noise study supporting scottishexecutive strategic environmental assessment for marine renewables. Tech.Rep. QinetiQ Ltd; 2007

[5] Wang D, Atlar M, Sampson R. An experimental investigation on cavitation,noise, and slipstream characteristics of ocean stream turbines. Proc Inst MechEng Part J Power Energy 2007;221(2):219e31. http://dx.doi.org/10.1243/09576509JPE310.

[6] Li Y, Çalisal SM. Numerical analysis of the characteristics of vertical axis tidalcurrent turbines. Renew Energy 2010;35(2):435e42. http://dx.doi.org/10.1016/j.renene.2009.05.024.

[7] ITTC. Cavitation committee report. In: 18th International Towing Tank Con-ference. 18the24th October, Kobe; 1987.

[8] Fleig O, Iida M, Arakawa C. Wind turbine blade tip flow and noise predictionby large-eddy simulation. J Sol Energy Eng 2004;126(4):1017e24. http://dx.doi.org/10.1115/1.1800551.

[9] Sezer-Uzol N, Long LN. 3-D time-accurate CFD simulations of wind turbinerotor flow fields. In: Proceedings of the 44th AIAA Aerospace Sciences Meetingand Exhibit. 9th-12th January, Reno; 2006. pp. 1e23.

[10] Tadamasa A, Zangeneh M. Numerical prediction of wind turbine noise. RenewEnergy 2011;36(7):1902e12. http://dx.doi.org/10.1016/j.renene.2010.11.036.

[11] McCann GN. Tidal current turbine fatigue loading sensitivity to waves andturbulence a parametric study. In: Proceedings of the 7th European Wave andTidal Energy Conference. 11the13th September, Porto; 2007.

[12] Nicholls-Lee R, Turnock S, Boyd S. Application of bend-twist coupled bladesfor horizontal axis tidal turbines. Renew Energy 2013;50:541e50. http://dx.doi.org/10.1016/j.renene.2012.06.043.

[13] Turnock SR, Phillips AB, Banks J, Nicholls-Lee R. Modelling tidal current tur-bine wakes using a coupled RANS-BEMT approach as a tool for analysingpower capture of arrays of turbines. Ocean Eng 2011;38(11e12):1300e7.http://dx.doi.org/10.1016/j.oceaneng.2011.05.018.

[14] McNaughton J, Rolfo S, Apsley DD, Stallard T, Stansby PK. CFD power and loadprediction on a 1MW tidal stream turbine with typical velocity profiles fromthe EMEC test site. In: Proceedings of the 10th European Wave and TidalEnergy Conference. 2nde5th September, Aalborg; 2013.

[15] Gant S, Stallard T. Modelling a tidal turbine in unsteady flow. In: Proceedingsof the 18th International Ocean and Polar Engineering Conference. Vancou-ver; 2008. pp. 473e9.

[16] Churchfield MJ, Li Y, Moriarty PJ. A large-eddy simulation study of wakepropagation and power production in an array of tidal-current turbines.PhilosTrans Ser Math Phys Eng Sci 2013;371(20120421):1e15. http://dx.doi.org/10.1098/rsta.2012.0421.

[17] Afgan I, McNaughton J, Rolfo S, Apsley D, Stallard T, Stansby P. Turbulent flowand loading on a tidal stream turbine by LES and RANS. Int J Heat Fluid Flow2013;43:96e108. http://dx.doi.org/10.1016/j.ijheatfluidflow.2013.03.010.

[18] Lloyd T, Turnock S, Humphrey V. Modelling techniques for underwater noisegenerated by tidal turbines in shallow waters. In: Proceedings of the 30thInternational Conference on Ocean, Offshore and Arctic Engineering.19the24th June, Rotterdam; 2011.

[19] Lloyd T, Humphrey V, Turnock S. Noise modelling of tidal turbine arrays forenvironmental impact assessment. In: Proceedings of the 9th European Waveand Tidal Energy Conference. 5th-9th September, Southampton, UK; 2011.

[20] Blake W. Aero-hydroacoustics for ships. Tech. Rep.vol. II. Bethesda: David W.Taylor Naval Ship Research and Development Center; 1984

[21] Moreau S, Roger M. Competing broadband noise mechanisms in low-speedaxial fans. AIAA J 2007;45(1):48e57. http://dx.doi.org/10.2514/1.14583.

[22] Lloyd TP, Turnock SR, Humphrey VF. Computation of inflow turbulence noiseof a tidal turbine. In: Proceedings of the 10th European Wave and Tidal EnergyConference. 2nde5th September, Aalborg; 2013.

[23] Sagaut P. Large eddy simulation for incompressible flows. Berlin: Springer-Verlag; 2006, ISBN 978-3540263449.

[24] Zang Y, Street R, Koseff J. A dynamic mixed subgrid-scale model and itsapplication to turbulent recirculating flows. Phys Fluids Fluid Dyn 1993;5(12):3186e96. http://dx.doi.org/10.1063/1.858675.

[25] James M, Lloyd T. Large eddy simulations of circular cylinders at a range ofReynolds numbers. In: Proceedings of ITTC Workshop of Wave Run-up andVortex Shedding. 17the18th October, Nantes; 2013.

[26] Spalding DB. A single formula for the “Law of the Wall”. J Appl Mech1961;28(3):455e8. http://dx.doi.org/10.1115/1.3641728.

[27] Kim Y, Castro IP, Xie ZT. Divergence-free turbulence inflow conditions forlarge-eddy simulations with incompressible flow solvers. Comput Fluids2013;84:56e68. http://dx.doi.org/10.1016/j.compfluid.2013.06.001.

[28] Lloyd TP. Large eddy simulations of inflow turbulence noise: application totidal turbines. Ph.D. thesis. University of Southampton; 2013.

[29] Issa R. Solution of the implicitly discretised fluid flow equations by operator-splitting. J Comput Phys 1986;62(1):40e65. http://dx.doi.org/10.1016/0021-9991(86)90099-9.

[30] Ffowcs Williams J, Hawkings D. Sound generation by turbulence and surfacesin arbitrary motion. Philos Trans Royal Soc Math Phys Eng Sci1969;264(1151):321e42. http://dx.doi.org/10.1098/rsta.1969.0031.

[31] Makarewicz R. Is a wind turbine a point source? J Acoust Soc Am 2011;129(2):579e81. http://dx.doi.org/10.1121/1.3514426.

[32] Welch P. The use of fast Fourier transform for the estimation of power spectra.IEEE Trans Audio Electroacoust 1967;15(2):70e3. http://dx.doi.org/10.1109/TAU.1967.1161901.

[33] Bahaj AS, Molland AF, Chaplin JR, Batten W. Power and thrust measurementsof marine current turbines under various hydrodynamic flow conditions in acavitation tunnel and a towing tank. Renew Energy 2007;32(3):407e26.http://dx.doi.org/10.1016/j.renene.2006.01.012.

[34] Walker JM, Flack KA, Lust EE, Schultz MP, Luznik L. Experimental and nu-merical studies of blade roughness and fouling on marine current turbineperformance. Renew Energy 2014;66:257e67. http://dx.doi.org/10.1016/j.renene.2013.12.012.

[35] Pope SB. Turbulent flows. Cambridge: Cambridge University Press; 2000, ISBN978-0521598866.

[36] Carolus TH, Schneider M, Reese H. Axial flow fan broad-band noise and pre-diction. J Sound Vib 2007;300(1e2):50e70. http://dx.doi.org/10.1016/j.jsv.2006.07.025.

[37] Bensow RE. Simulation of unsteady propeller loads using OpenFOAM. In:Proceedings of the 16th Numerical Towing Tank Symposium. 2nd-4thSeptember, Mülheim; 2013.

[38] McNaughton J, Afgan I, Apsley DD, Rolfo S, Stallard T, Stansby PK. A simplesliding-mesh interface procedure and its application to the CFD simulation ofa tidal-stream turbine. Int J Numer Methods Fluids 2013. http://dx.doi.org/10.1002/fld.3849.

[39] Banks J, Bercin K, Lloyd TP, Turnock SR. Fluid structure interaction analyses oftidal turbines. In: Proceedings of the 16th Numerical Towing Tank Sympo-sium. 2nde4th September, Mülheim; 2013.

[40] Banks J. Modelling the propelled resistance of a freestyle swimmer usingComputational Fluid Dynamics. Ph.D. thesis. University of Southampton;2013.

[41] Jiang CW, Chang M, Liu Y. The effect of turbulence ingestion on propellerbroadband forces. In: Proceedings of 19th Symposium on Naval Hydrody-namics. 23rde28th August, Seoul; 1994. pp. 751e69.

[42] Oerlemans S, Sijtsma P, Mendezlopez B. Location and quantification of noisesources on a wind turbine. J Sound Vib 2007;299(4e5):869e83. http://dx.doi.org/10.1016/j.jsv.2006.07.032.

[43] Migliore P, Oerlemans S. Wind tunnel aeroacoustic tests of six airfoils for useon small wind turbines. J Sol Energy Eng 2004;126(4):974e85. http://dx.doi.org/10.1115/1.1790535.

[44] Powell A. Theory of vortex sound. J Acoust Soc Am 1964;36(1):177e95. http://dx.doi.org/10.1121/1.1918931.

[45] Morton M, Devenport W, Alexander WN, Glegg SAL, Borgoltz A. Rotor inflownoise caused by a boundary layer: inflow measurements and noise pre-dictions. In: Proceedings of the 18th AIAA/CEAS Aeroacoustics Conference.4th-6th June, Colorado Springs; 2012. http://dx.doi.org/10.2514/6.2012-2120.

[46] Goldstein ME. Aeroacoustics. New York: McGraw-Hill; 1976, ISBN 0-07-023685-2.

[47] Howe MS. Acoustics of fluid-structure interactions. Cambridge: CambridgeUniversity Press; 1998, ISBN 0-521-63320-6.

[48] Betschart M. Andritz Hydro Hammerfest. In: Ocean Renewable Energy GroupAnnual Conference. 13th-14th September, Halifax; 2012. http://www.marinerenewables.ca/wp-content/uploads/2012/09/Michael-Betschart-OREG-2012.pdf.

[49] Molland AF, Bahaj AS, Chaplin JR, Batten WMJ. Measurements and predictionsof forces, pressures and cavitation on 2-D sections suitable for marine currentturbines. Proc Inst Mech Eng Part J Eng Marit Environ 2004;218(2):127e38.http://dx.doi.org/10.1243/1475090041651412.

[50] Urick R. Principles of underwater sound for engineers. 3rd ed. New York:McGraw-Hill; 1996, ISBN 978-0932146625.

[51] StarzmannR, BaldusM,Groh E, HirschN, LangeNA, Scholl S. Full-scale testing ofa tidal energy converter using a tug boat. In: Proceedings of the 10th EuropeanWave and Tidal Energy Conference. 2nd-5th September, Aalborg; 2013.