1 Introduction

With the fast development of the wind turbine industry, wind farm is becoming closer and closer to the residential area, and the size of wind turbine blade is becoming larger and larger. Therefore, the wind turbine aerodynamic noise becomes more and more serious. A lot of complaints are made to environmental protection organization and then force many countries to release strict regulations on the wind turbine noise. As a result, the wind turbine aerodynamic noise has been a main issue that restricts the wind turbine development. Therefore, effective wind turbine low aerodynamic noise design and suppression techniques are urgently required. However, the generation mechanisms behind the wind turbine aerodynamic noise are still not fully clear, the accurate prediction is rather difficult, and the noise suppression techniques are in lack of theoretical basis. Therefore, deep understanding of the wind turbine noise generation mechanism, accurate prediction methods, and effective noise suppression techniques have received much attention recently by industry and academic^[1].

2 Sources of wind turbine aerodynamic noise

The wind turbine aerodynamic noise is generated by the interaction between the incoming flow and wind turbine blade, and can be categorized into three different generation mechanisms^[2], i.e., (a) low frequency noise, which includes the steady load noise by the rotating blade or lifting surface and unsteady load noise by the blade passing the flow deficit region or waking region; (b) inflow turbulence noise, a typical broadband noise, which is produced by the interaction between the inflow turbulence and the blade and is not so easy to measure due to the unstable inflow; (c) airfoil self-noise, which is a typical broadband noise and generated by the flow past the airfoil surface. The tonal noise can be observed due to the blunt trailing edge. The airfoil self-noise mainly consists of (i) trailing edge noise, which is generated by the interaction between the turbulent boundary layer and the trailing edge and is a broadband noise, (ii) airfoil tip noise, which is generated by the interaction between the turbulent flow around the tip and the blade tip surface and still needs much more research, (iii) separation noise, which is generated by the interaction between the turbulent and the blade and is a broadband noise, (iv) laminar boundary layer noise, which is generated by the nonlinear interaction between the boundary layer and the blade and is mainly a tonal noise, (v) blunt trailing edge noise, which is generated by the vortex shedding from the blunt trailing edge and is mainly a tonal noise, and (vi) noise by holes and slits, which is produced by the unsteady shear flow past the holes and slits and where the tonal noise dominates the spectrum.

3 Theoretical prediction of wind turbine aerodynamic noise

Theoretical prediction methods for wind turbine aerodynamic noise include the theoretical model and the semi-empirical prediction formulation. Theoretical models are based on the analytical solution of flow field around the blade and acoustics. In order to obtain the analytical solution, most of the theoretical models make some approximation on the basis of satisfying the basic physics laws. Combined with the theoretical analysis, a large set of experiment data have been obtained to establish the basic scaling principles. Then, the semi-empirical formulation is established by use of curve fitting and correlation analysis on the experiment measurements.

3.1 Theoretical models

Ffowcs Williams and Hall^[3] developed the trailing edge noise model for the semi-infinite zero-thickness plate, in which they introduced quadruple source and Green's function for the semi-infinite plate to obtain the solution of the Lighthill theory^[4]. Crighton^[5] obtained another theoretical model based on the model of a series of vortex past the semi-infinite plate trailing edge. Amiet^[6] correlated the far field pressure with the pressure on the wall in the upstream of the trailing edge directly. Howe^[7] obtained a low Mach number trailing edge noise model in 2001 by use of the solution of the finite-chord Green's function. To avoid the calculation of Green's function, Amiet^[6] obtained the radiated noise from the solution of pressure wave on the wall by solving the scattering problem of equivalent wave based on the assumption of semi-infinite airfoil. Roger and Moreau^[8] obtained a low Mach number trailing edge noise model which considered the scattering effects of leading edge based on the correction of radiation integration parameters^[6] when the mean flow convection was omitted in 2004.

3.2 Semi-empirical formulation

The typical semi-empirical formulation in the airfoil self-noise is the Brooks-Pope-Marcolini (BPM) formula developed by Brooks et al.^[9] based on theoretical models of Ffowcs Williams and Hall^[3] and Amiet^[10]. The BPM semi-empirical method gives the formulation for five different noise sources. Paterson et al.^[11] found the empirical relationship between the tonal frequency and flow incoming velocity based on the NACA0012 experiments. Fink^[12] obtained the airfoil trailing edge noise semi-empirical formulation based on the correlation analysis of a large set of experiment data. Glegg et al.^[13] proposed a wind turbine broadband noise prediction method based on the experiment measurements of wind turbine broadband noise. Zhu et al.^[14] improved the prediction accuracy of tip vortex noise based on a new tip vortex correction technique. Bai and Li^[15] analyzed the turbulent boundary layer trailing edge noise in the traditional BPM method and found that the traditional BPM of the turbulent boundary layer trailing edge noise over predicted the contribution from pressure side sources. Then, they further improved the traditional BPM, and obtained the predicted results with high accuracy for the high angles of attack and thick airfoil.

Grosveld^[16] proposed a full-scale wind turbine semi-empirical prediction formulation considering the noise from inflow turbulence, turbulent boundary layer and blunt trailing edge wake, and applied the method to the real wind turbine to predict the radiated noise. The predicted results matched very well with the experiment data of MOD-OA, MOD-2, and WTS-4. Kim et al.^[17] further developed a wind turbine aerodynamic noise prediction method. Due to the introduction of the nonlinear composite beam theory to the wind turbine aeroacoustic prediction method, this method can account for the fluid-structure interaction of wind turbine. Li et al.^[18] applied the semi-empirical formulation on the noise research of the AOC-15/50 wind turbine.

The theoretical models for the wind turbine noise can assess the wind turbine noise quickly. Therefore, it is used widely in industry. However, many approximation and assumptions, are made on the theoretical models, and the theoretical models lack a detailed source model. Therefore, it cannot quantitatively predict the real and complex wind turbine noise and design the low noise wind turbine.

4 Experimental measurements of wind turbine aerodynamic noise

The experimental methods of wind turbine noise mainly include flow field measurements, noise source localization techniques, and far field noise measurements. For the flow field measurement, the commonly used instruments are the Pitot tube, hot wire anemometry, hot film anemometry, laser Doppler velocimeter (LDV), and particle image velocimetry (PIV). The working part of Pitot tube and hotwire/hot film anemometry needs to be located into the flow field and will disturb the flow field. The LDV is the technique to measure the flow velocity by using the reflection, refraction and transmission of moving particle and can be placed outside of flow being measured. Therefore, it has no effect on the flow field and has high measurement accuracy. However, the LDV cannot measure the high speed moving particle because of the limited frequency difference of two-frequency laser. The PIV is an optical method of flow visualization. The fluid is seeded with tracer particles, and the spatial flow structure and characteristics can be studied by the motion of the seeding particles. The method of the PIV is, to a large degree, nonintrusive. The added tracers (if they are properly chosen) generally cause negligible distortion of the fluid flow. Therefore, it can reveal the real flow field information. Noise source localization reconstructs the source frequency, location and power information by using the beamforming algorithm based on the microphone array. Far field noise measurements place the microphones at different far field locations to measure the wind turbine noise spectrum for the investigation of characteristics of sound source.

Many researches have been conducted for wind turbine self-noise. Tam^[19] proposed the acoustic feedback mechanism in the wind turbine trailing edge noise, which led to many tonal noise based on the analysis of a large set of experiment data in the 1970s. Up to now, many different versions of the acoustic feedback loop generation mechanism have been developed, and the one that has received acceptance by most acousticians thinks that noise is generated at the trailing edge, and then sound waves propagate upstream and disturb the boundary layer to generate a Tollmien-Schlichting (T-S) instability wave. The T-S instability wave emits noise outside at the trailing edge, and the tonal noise is generated. Paterson et al.^[11] conducted the far field noise measurement to the airfoil vortex shedding noise at the low turbulence incoming flow and found that the vortex shedding noise was mainly the tonal noise. Roger and Moreau^[8] and Arbey and Bataille^[20] found that the Reynolds number, the angle of attack, and the trailing edge played a role in generation of trailing edge noise based on the far field microphone measurement. For example, in the large Reynolds number ($>10^6$), the trailing edge noise spectrum is mainly broadband. When the incoming flow is two dimensional and at the low Reynolds number, the high frequency tonal noise dominates the spectrum, which is radiated as sound into the far field by the instability wave (for example, the T-S wave) in the boundary layer at the trailing edge. Brooks et al.^[9] conducted the experiment measurement for the airfoil aerodynamic noise by multiple microphones and found that the airfoil flow-induced noise was mainly a broadband noise with tonal noise. Moroz^[21] found that the blunt trailing edge could produce the tonal noise after the measurement of a diameter 7.6 m wind turbine. Oerlemans et al.^[22] applied the phased microphone array technique on the noise measurement of the 58 m three-rotor wind turbine and found that the wind turbine noise was mainly from the broadband trailing noise. Kameier and Neise^[23-24] revealed experimentally that the instability of rotor had a great effect on the tip noise. Jacob et al.^[25] carried out near and far field aeroacoustic measurements about the tip leakage flow past a single non-rotating airfoil. Wang et al.^[26] measured the small-scale wind turbine blade tip near the wake noise radiation and showed that the noise radiation spectrum of rotor blade tip was composed of the broadband noise superposed by the discrete noise from the fundamental frequency and its harmonic in the rotating process of the rotor.

For the inflow turbulence noise research, Marcus and Harris^[27] conducted experiments to study the impulsive noise of a horizontal axis wind turbine and found that the impulsive noise resulted from the repeated blade passage through the mean velocity deficit induced in the lee of wind turbine support tower. The two factors which most influence this noise are the rotation speed and the tower drag coefficient. Jacob et al.^[28] used the PIV technique to measure the rod/NACA0012 airfoil unsteady flow field and far field noise, and showed that the interaction of vortex and leading edge of airfoil was the major noise source. Rogers and Omer^[29] investigated the effect of the incoming flow turbulent intensity on the aerodynamic noise, and found that with the increase in the incoming flow turbulent intensity, the wind turbine aerodynamic noise increased significantly.

For the research of wind turbine low frequency noise, Jakobsen^[30] conducted an experiment to further study the wind turbine low frequency noise, and found that wind turbines with a downwind rotor generated much stronger low frequency noise due to the interaction of shedding vortex from supporting tower and wind turbine. Later, Leventhall^[31] investigated the wind turbine low frequency noise and came to the same conclusion that the interaction of shedding vortex and wind turbine was the major low frequency noise source. Jung et al.^[32] studied experimentally the infrasound and low frequency noise of large wind turbines, and found that, with the increase in the size of wind turbines, the low frequency noise level increased as well.

5 Numerical simulation of wind turbine noise

The numerical simulation method of wind turbine noise mainly includes the hybrid method and the direct numerical computation. With the hybrid method, the near field aerodynamics is calculated to obtain the pressure load and turbulent fluctuations that form the acoustic source terms for a separate calculation of the far field acoustics. For both the simulation of the flow and the computation of the sound waves, there are a variety of methods that differ in accuracy and demand for computational resources. The simulation of flow field near the source region can be conducted by the unsteady Reynolds averaged Navier-Stokes (URANS) method, the large eddy simulation(LES) and so on. Far field acoustics can be obtained not only by solving the FW-H equation^[33] and the Kirchhoff integral^[34] but also by solving the Euler or wave equation in a larger domain. The hybrid method is suitable for the research of noise generation mechanism because it can cover more real physics than the approaches of the theoretical model and semi-empirical formula.

With the development of computational aeroacoustics (CAA), the direct numerical computation becomes a more promising method for investigating the noise generation mechanism. This is because the direct method does not use any simplification in the computation that is more or less involved in other approaches. With the direct method, the process of noise generation and propagation and the interaction of flow and acoustics can be reproduced accurately.

For the wind turbine blade self-noise, Singer et al.^[35] investigated the characteristics of trailing edge noise by the FW-H integration in 2000. In their research, the flow field over the blade was calculated by the computational fluid dynamics (CFD). Lummer et al.^[36] studied the interaction between the vortex shedding and the trailing edge and revealed that the strength of interaction between the nonlinear vortex and the trailing edge was higher than that of the linear vortex. Jiang et al.^[37-38] conducted the high-order numerical simulation of airfoil self-noise by the CAA method. The tonal noise was found to move to the low frequency range and finally become broadband with the increase in the angle of attack. Fleig and Arakawa^[39] performed the LES on the flow over the wind turbine blade and predicted the far field broadband noise by the FW-H integration. The tip vortex noise of wind turbine blade was investigated by Marsden et al.^[40] and Iida et al.^[41]. Bai and Li^[42] developed a RANS-based airfoil turbulent trailing edge noise prediction method with a very good agreement between the predicted results and the experiment data.

The incoming turbulence interaction noise is another important research topic for the investigation of wind turbine noise. In 2004, Morris et al.^[43] analyzed the effects of gust, turbulence and wind shear on the aerodynamic noise of wind turbine by the CAA method. Jacob et al.^[25] and Greschner et al.^[44] predicted the interaction noise of rod and NACA0012 airfoil by the hybrid method, and indicated that the combination of the detached eddy simulation (DES) with the FW-H equation was very suitable for computing the broadband noise of complicated flow. Meanwhile, Jiang et al.^[45] used the numerical simulation to study the interaction noise of airfoil and incoming flow. As to the overall noise of wind turbine, Tadamasa and Zangeneh^[46] computed the far field noise of axis wind turbine by using the commercial CFD software.

In general, the numerical simulation is able to provide a detailed description of noise generation and propagation. However, the wind turbine noise is a typical multi-scale problem that still remains a big challenge for the numerical simulation due to the high demand for the computational resources. With the development of numerical schemes and computers, the numerical simulation will become one of the most important methods for the investigation of wind turbine aerodynamic noise.

6 Noise suppression techniques of wind turbine

The noise suppression techniques of wind turbine are always the focus of wind turbine noise researches. In 1972, Roger and Robert^[47] studied the effect of leading edge serrations on the flow-induced noise radiation. Bohn^[48] conducted a noise experimental measurement for the airfoil with a porous flow-wise extension and found that the noise reduction from airfoil with the porous extension was frequency dependent. Geyer et al.^[49] studied the influence of porous airfoil to the aerodynamic noise by making use of phased microphone array, and found that the porous wall could reduce the trailing edge noise significantly. Wind tunnel experiments by Fink and Bailey^[50] and European reduction of airframe and installation noise (RAIN) program proved that the porous edge extension, such as the trailing edge with serration and brushes, was an effective trailing edge noise reduction technique. Howe^[51] derived the noise prediction formula for sawtooth trailing edge. Chong et al.^[52] used the non-flat plate type trailing edge serrations to reduce the airfoil trailing edge self-noise, and found that the non-flat plate serration not only reduced the broadband self-noise significantly, but also eliminated the high-frequency noise that was observed by others who used the flat plate type serration. Xu and Li^[53] and Xu et al.^[54] studied the trailing edge noise of airfoil with serration and brushes experimentally, and found that the serrated trailing edge and trailing edge with brushes could reduce the airfoil self-noise significantly. Glezer^[55] introduced synthetic-jet actuation on the axis wind turbine blade to control the stall flow and found that it could reduce the airfoil self-noise. Greenblatt et al.^[56] reduced the airfoil self-noise using plasma actuators in the leading edge of the axis wind turbine to control the separation.

Wind turbine aerodynamic noise suppression techniques are strongly relied on the research of generation mechanism of flow-induced noise. Currently, the generation mechanism is still unclear. Therefore, efficient noise reduction techniques require further indepth researches.

7 Summary and outlook

The paper summarizes the research status of wind turbine aerodynamic noise. The main components of wind turbine aerodynamic noise are introduced with the specific attention to the application of theoretical prediction, experiment measurements, and numerical simulation methods on the wind turbine noise and its existing problem. The current status and disadvantage of the wind turbine noise suppression techniques are analyzed.

The current available wind turbine aerodynamic noise researches mainly focus on the noise from small and the middle-sized wind turbine, in which the noise is in the middle or high frequency range. With the size of wind turbine becoming larger and larger, the future wind turbine noise dominates in the low frequency range, which requires more researches on the wind turbine low frequency noise generation mechanism and reduction techniques. The theoretical prediction, experimental measurements, and numerical simulation method are highly desired for the wind turbine low frequency noise prediction and reduction.