附錄翻譯原文Coarse-resolution numerical prediction of small wind turbine noise with validation against field measurementsAbstractThe noise emission and the power output from a small horizontal axis wind turbine is investigated using coarse-resolution computational fluid dynamics (CFD) simulations conducted with the commercial software STAR-CCM+?. The steady Reynolds-averaged Navier-Stokes (RANS) and transient delayed detached-eddy simulation (DDES) methodologies were used for the prediction of the flow field around the wind turbine. It is found that the DDES method with the Spalart-Allmaras turbulence model provides predictions of the wind turbine power that are in good conformance with available field measurements. The aeroacoustic calculations were performed using both the STAR-CCM+? acoustic model and an in-house code. The in-house code implemented both the permeable and impermeable formulations of the Ffowcs Williams and Hawkings (FW-H) equation. The predicted A-weighted sound pressure level (SPL) spectra, as well as the apparent SPL, obtained from the permeable formulation of the FW-H equation agree well with the wind turbine acoustic field measurements. It is found that the presence of the tower slightly decreases the wind turbine power output at all simulated incident wind speeds. It is also found that the presence of the tower leads to modifications of the SPL spectra at frequencies between about 300 and 1500 Hz.Keywords:Wind turbine noise; Wind turbine power; Wind turbine field measurements; CFD simulation; FW-H equationNomenclaturec Speed of sound unFluid velocity in the normal direction to the bodyD Wind turbine rotor diameter ut Shear velocityDmPoint of intersection of the viscous and fully turbulent viComponent of local velocity on the bodyregions of the flow vnLocal velocity on the body in the direction normal toE Log-law coefficient the body surfaceE0 Modified log-law coefficient vrefReference velocity for power-law inlet boundaryf Roughness function condition prescriptionfv3 Damping function vWTVelocity prescribed at the inlet boundary for windG Free-space Green’s function turbine simulationH(f) Heaviside (step) function xob Observation position vectork Turbulence kinetic energy yt Dimensionless wall distanceLAeq,k Background corrected A-weighted SPL ymtPoint of intersection of the viscous and fully turbulentLiComponent of vector defined in Equation (7) flow regions eDm ? ymtTLM ?LiMi yn Normal distance from the wall to the wall-cell centroidLr ?Liri yrefReference height for power-law inlet boundaryLWA,k Apparent sound power level in dBA condition prescriptionMMagnitude of local Mach number vector of source ys Source position vectorMiComponent of local Mach number vector of source yWTCoordinates in the wall-normal (y) direction at thedefined in Equation (7) inlet boundaryMrMach number of source in the direction of radiation, eMiri Generalized functionniComponent of unit outward normal vector to surface _ Source-time differentiationDPij Local force intensity L D’Alembertian operatorp0 Gauge pressure V2 Laplacian operatorR1Slant distance from the rotor center to the microphone unFluid velocity in the normal direction to the bodylocation ut Shear velocityr Distance between observer and source viComponent of local velocity on the bodyriComponent of unit vector in the direction of radiation vnLocal velocity on the body in the direction normal to~ the body surfaceS Deformation parameter vrefReference velocity for power-law inlet boundaryS0 Reference area S0 ? 1 m2 condition prescriptionTij Lighthill stress tensor vWTVelocity prescribed at the inlet boundary for windt Observation time turbine simulationUiComponents of vector defined in Equation (7) xob Observation position vectorUn ?Uini yt Dimensionless wall distanceUs Streamwise velocity ymtPoint of intersection of the viscous and fully turbulentut Dimensionless velocity flow regions eDm ? ymtTui Component of local fluid velocity ynNormal distance from the wall to the wall-cell centroidGreek lettersKronecker delta function Dirac delta function Turbulence dissipation rate von Karman constant Kinematic fluid viscosity Density of fluidDensity perturbation, r r0 Source timeSubscripts0 Fluid variable in quiescent medium L Loading noise componentT Thickness noise component1. IntroductionClimate change concerns, as well as increasingly aggressive efforts from government to develop the policy and action to mitigate these changes, have led to the increasing development of a wide spectrum of innovative energy technologies for the efficient utilization of various sources of renewable energy. These renewable energy technologies include solar power [1], thermoelectricity [2], wind energy, geothermal energy, tidal energy, and bio energy (biomass and biofuels). The focus of this paper will be on wind energy, with a particular emphasis on the noise issues associated with the operation of wind turbines.The prediction of wind turbine noise is still a challenging problem for the computational aeroacoustics (CAA) community, owing to the large range of scales contained in the highly disturbed flow field surrounding a wind turbine. These include the largest scales associated with the vortex shedding from the support tower all the way down to the finest scales of the turbulence generated in the attached boundary layer of the wind turbine [3]. A direct simulation for both the flow and acoustic fields is not currently possible owing to the severe limitations of the computational resources. In consequence, many hybrid models (or schemas) have been proposed for the separate (decoupled) simulation of the wind turbine aerodynamic and acoustic fields. These can be classified into two main categories: namely, the semi-empirical and the computational-based methods, as summarized in Table 1.Semi-empirical models for aerodynamic and aeroacoustic calculations associated with wind turbines are described in Refs. [12] ; [13]. These semi-empirical aerodynamic models are based on the blade element method (BEM) [17]. The main idea underpinning the BEM is to analyze the wind turbine flow field by dividing the wind turbine blade into a number of independent elements and calculating the aerodynamic forces on each of these elements using tabulated airfoil data, which are obtained from wind tunnel measurements that have been subsequently corrected for three-dimensional effects. The inputs for the BEM-based model are generally the blade geometric parameters and the wind turbine operating conditions. The outputs from the model are the local Reynolds number, the local angle of attack (AOA) and the boundary-layer displacement thickness, which can be used as the relevant inputs to a semi-empirical noise prediction model. As in the case of the BEM, the semi-empirical noise prediction models also divide the wind turbine blade into segments and treat each of these segments as two-dimensional airfoil sections, each acting as an incoherent sound source [13]. The most common noise prediction models of this type are the Brooks, Pope and Marcolini (BPM) model [18], the TNO model [19], and the models proposed by Amiet [20] and by Lowson [21]. These noise prediction models are usually employed in conjunction with noise propagation models. Based on sound ray theory, these noise propagation models provide a set of semi-empirical formulae that account for various effects (e.g., air absorption, terrain, temperature gradients) on the sound propagation. Owing to their simplicity and ease of calculation, this set of semi-empirical methodologies for the wind turbine noise prediction is by far the most widely applied method for engineering applications (for both large modern and small-sized wind turbines). Table 1 lists some of the references that describe the use and application of semi-empirical methods for wind turbine noise prediction.Despite its high computational efficiency, semi-empirical methods do not provide the detailed and unsteady flow field information that is required for an accurate noise prediction for wind turbines. As the computational technology continues to advance, detailed flow field information in the vicinity of a wind turbine can be obtained from computational fluid dynamics (CFD) based simulations for realistic problems. Table 1 provides some key references for the second category of wind turbine noise prediction methods, which apply physics-based computational models for both the aerodynamic and aeroacoustic calculations. Owing to the high computational requirements, this second category of methods can only be used for simulating a portion of the wind turbine, and few researchers have used the computational schema for wind turbine noise prediction to date. Filios et al. [15] used the three-dimensional (3D) low-order panel method with a boundary-layer correction model and the impermeable Ffowcs Williams and Hawkings (FW-H) formulation for the prediction of noise for the National Renewable Energy Laboratory (NREL) Phase II downwind arranged wind turbine. In this study, relatively good predictive agreement was achieved for the aerodynamic data, in spite of the fact that the simulation only included the wind turbine rotor. Furthermore, no acoustic measurements were presented for the validation of the noise predictions made in this study.A more computational demanding flow solver was used by Zhu [16] for the NREL 5-MW horizontal axis wind turbine (HAWT) noise prediction. The incompressible/compressible splitting method, proposed by Hardin and Pope [22], was applied in this study. Furthermore, the flow field prediction was obtained using the incompressible Reynolds-averaged Navier-Stokes (RANS) methodology and the associated acoustic field was determined using the inviscid acoustic equations in the modified form proposed by Shen and S?rensen [23]. High-order spatial and temporal discretization schemes, dispersion-relation-preserving (DRP) and classical Runge-Kutta numerical techniques were employed to solve these acoustic equations. Owing to the tremendous number of grid points (320 million) and the very small time step (1.4 × 10?7 s) used, only one blade was simulated for a total blade rotation of 20.4° in this study. The simulation was carried out using 112 processors and the authors reported that the simulations that they conducted had already reached the limits of the computational capabilities available to them.By using a coarse mesh and a less computationally demanding noise prediction model (namely, the acoustic analogy method), Tadamasa and Zangeneh [14] conducted a RANS simulation for the noise produced by a wind turbine with the commercial CFD software ANSYS CFX?. These simulations were conducted for a single blade of the NREL Phase VI two-bladed wind turbine. Both the permeable and impermeable FW-H formulations were employed for the noise calculations. The blade pressure results were shown to agree relatively well with some measurements, but the sound pressure level (SPL) results were found to decrease smoothly over the whole spectrum of frequencies. This might be due to the time-averaging of the quantities implicit in the RANS model, which filters out all the small velocity fluctuations that are expected to contribute to the broadband noise. Unfortunately, none of currently available investigations using computational models include the wind turbine tower in their simulations, owing to the limited computational resources in these studies. Furthermore, these investigations also do not compare their numerical results with any wind turbine noise measurements. Finally, no noise predictions from a full-sized wind turbine have been undertaken to date. The lack of available wind turbine noise data has severely limited the validation of the numerical predictions for the noise.In view of this, the objective of this paper is to present aerodynamic and acoustic predictions (less than 1000 Hz) for a small commercial HAWT and to validate the aerodynamic and aeroacoustic models using field measurements of power and sound pressure levels that have been made for this small HAWT. Due to the computational limitations, a relatively coarse mesh was employed for the aerodynamic flow simulations. These simulations were performed using the commercial CFD software STAR-CCM+?. The acoustic field was calculated using the STAR-CCM+? acoustic model and an in-house code implementing both the FW-H permeable and impermeable formulations. Rotor only and a full wind turbine (which includes the tower) configurations have been simulated in the current study in order to evaluate the aerodynamic and aeroacoustic effects of the tower. Different combinations of the CFD and acoustic simulations were also examined and evaluated with reference to the available field measurement (power and noise) data.2. Field measurementsThe measurements of power and sound pressure level were carried out on a three-bladed WINPhase 10 small upwind arranged wind turbine. The WINPhase 10 wind turbine has a rated output power of 10 kW with a rotor speed of 150 rpm at a wind speed of 11 m s?1. The diameter of the rotor for this small wind turbine is 10 m and the height of the tower is 20 m. We note that the field measurement report is an unpublished contract report made available to the authors by WINPhase Energy Inc. The limited measurement details available will be described below, as well as the final measurement results extracted from this contract report.The field measurements were conducted on the wind farm Sunite over a period of two days. This wind farm is located in the province of Inner Mongolia (northern part of China). The measurements were performed in accordance to the recommendations provided by the American Wind Energy Association (AWEA) [24] and International Electrotechnical Commission standards [25]. One BSWA MP201 microphone was placed on an acoustic hardboard to measure the 10-s averaged A-weighted SPL at a reference location. The non-acoustic data in the experiment were acquired by the data acquisition system at a sampling rate of 2 Hz.The measured A-weighted SPL was averaged and separated from the background noise for each one-third octave band for wind speeds in the range from 4 to 11 m s?1. The resulting A-weighted SPL for each one-third octave band for each wind speed was summarized in the field measurement report [26]. For the cases where the background noise was within 3 dBA of the overall noise, no values for the A-weighted SPL were reported. The apparent sound power level in dBA [24] was calculated from 21046WAKeqRLS??????????,,logwhere LAeq,k is the background corrected A-weighted SPL under the reference conditions, R1 is the slant distance from the rotor center to the microphone location, and S0 is the reference area (S0 = 1 m2). The ?6(dBA) term in Equation (1) accounts for the effects of the hardboard used in the acoustic measurements.3. Theoretical framework3.1. CFD methodologyThe steady RANS equations, applied in conjunction with the Spalart-Allmaras (S-A) and the two-layer k?ε turbulence closure models, were solved to provide predictions for the wind turbine power. The transient delayed detached-eddy simulation (DDES) methodology was also used with the S-A turbulence closure model for wind turbine power predictions, as well as for providing the sound source information for the wind turbine noise predictions.The one equation S-A model implemented in STAR-CCM+? is the standard S-A turbulence model [27] with rotation/curvature corrections [28] and with the fv3 damping function added to the deformation parameter S? term [29]. The k?ε two-layer model used for the RANS simulation applies the one-equation Wolfstein model [30] close to the wall and the two-equation k?ε model away from the wall [31].An all-range y+ wall treatment was used in STAR-CCM+? for both the S-A and the k?ε two-layer turbulence closure models. In the region near the wall, the normalized streamwise velocity profile in the viscous and turbulent boundary layers are given by (u+=defUs/uτ where uτ is the shear velocity and Us is the streamwise velocity)where y+=defuτyn/ ν (v is the kinematic fluid viscosity) andHere, ym+ is the point where the viscous sublayer intersects the log-law layer. This point is determined using Newtonian iteration. The parameter κ is the so-called von Karman constant (κ = 0.41) and f is the roughness function which modifies the log-law coefficient E [32]. This formulation for the wall treatment uses Reichardt’s law [33] for the calculation of the source term in the discretization of the momentum transport equation for those discretization cells that abut against the solid wall. For a coarse mesh where the mesh point closest to the wall lies in the region y+≥30, the term in the square brackets in Equation (2) tends to unity and Equation (2) in this case reduces to the standard logarithmic law of the wall (referred to as the high y+ wall treatment). Alternatively, for a fine mesh where the mesh point closest to the wall lies in the region y+ ≈ 1, Equation (2) resolves the properties of the flow all the way down to the wall (referred to as the low y+ wall treatment). When y+ falls within the buffer layer, this method provides more realistic predictions of the flow than either the low y+ or the high y+ wall treatments.3.2. CAA frameworkThe original FW-H formulation was published in 1969 [34]. In this formulation, generalized functions were utilized to recast the continuity and momentum equations into the form of an inhomogeneous wave equation with monopole and dipole sources on the body surface and a quadrupole source distribution in the volume surrounding the body as follows:where ?? is the d’Alembertian operator defined as ?=1c2?ˉ2?t2??ˉ2, c is the speed of sound, vn is the local velocity at a location on the body in the direction normal to the body surface implicitly defined by f (xob,t) = 0, and un is the fluid velocity at the same location in the normal direction to the body. Furthermore, ΔPij is the local force intensity. In general, the viscous term in ΔPij is negligible [35], so ΔPij = p′δij where δij is the Kronecker delta function and p′ is the gauge pressure. Finally, Tij is the Lighthill stress tensor defined as Tij=defρuiuj+pij? c2ρ’ where ρ′ = ρ?ρ0, and δ(f) and H(f) are the Dirac delta and Heaviside (step) functions, respectively. In Equation (4), the bars over the partial derivatives are used to denote a generalized partial derivative [36]. The term c2 (ρ? ρ 0) can be replaced by p′ in accordance to the linear wave propagation theory.To realize the advantages of the FW-H equation, an integral formulation of this equation can be obtained by convolving Equation (4) with the free-space Green’s function G = δ(g)/4 πr, where (r = |xob?ys|, g = τ?t + r/c ((xob,t) and (ys, τ) are the receptor (observation) and source space-time variables, respectively). For non-deformable surfaces, an integral formulation of the permeable FW-H equation can be obtained. The use of the descriptor “permeable” for this equation refers to the fact that the surface can be placed outside the solid body, allowing the fluid to flow through it. Considering only the surface terms in Equation (4), the permeable FW-H formulation assumes the following form:The dots over the quantities denote temporal derivatives with respect to the source time τ , and the subscript ret indicates that the quantity is evaluated at the retarded time When using the permeable FW-H formulation, the quadrupole sources enclosed within the integration surface contribute through the surface source terms in Equation (4). Any physical source of sound or propagation effect outside the integration surface contributes through the volume source term. For the wind turbine noise calculations, the flow Mach number (Ma) is relatively low (Ma ≈ 0.2), which implies a less efficient quadrupole noise source in the flow. In addition, the quadrupole source outside the integration surface is assumed t