Aeroacoustic airfoil shape optimization enhanced by autoencoders_2

2025-04-27 0 0 768KB 40 页 10玖币
侵权投诉
Aeroacoustic airfoil shape optimization enhanced by
autoencoders
Jiaqing Koua, Laura Botero-Bol´ıvarb, Rom´an Ballanoa, Oscar Marinoa,
Leandro de Santanab, Eusebio Valeroa,c, Esteban Ferrera,c
aETSIAE-UPM-School of Aeronautics, Universidad Polit´ecnica de Madrid, Plaza
Cardenal Cisneros 3, E-28040 Madrid, Spain
bDepartment of Thermal Fluid Engineering, University of Twente, PO Box 217, 7522
NB Enschede, The Netherlands
cCenter for Computational Simulation, Universidad Polit´ecnica de Madrid, Campus de
Montegancedo, Boadilla del Monte, 28660, Madrid, Spain
Abstract
We present a framework for airfoil shape optimization to reduce the trailing
edge noise for the design of wind turbine blades. Far-field noise is evaluated
using Amiet’s theory coupled with the TNO-Blake model to calculate the
wall pressure spectrum and fast turn-around XFOIL simulations to evalu-
ate the boundary layer parameters. The computational framework is first
validated using a NACA0012 airfoil at 0°angle of attack. Particle swarm
optimization is used to find the optimized airfoil configuration. The multi-
objective optimization minimizes the A-weighted overall sound pressure level
at various angles of attack, while ensuring enough lift and minimum drag.
We compare classic parametrization methods to define the airfoil geometry
(i.e., CST) to a machine learning method (i.e., a variational autoencoder).
We observe that variational autoencoders can represent a wide variety of ge-
ometries, with only four encoded variables, leading to efficient optimizations,
which result in improved optimal shapes. When compared to the baseline ge-
ometry, a NACA0012, the autoencoder-based optimized airfoil reduces by 3%
(1.75 dBA) the overall sound pressure level (with decreased noise across the
Email addresses: jiaqingkou@gmail.com (Jiaqing Kou),
l.boterobolivar@utwente.nl (Laura Botero-Bol´ıvar), r.ballanom@alumnos.upm.es
(Rom´an Ballano), oscar.marino@upm.es (Oscar Marino),
leandro.desantana@utwente.nl (Leandro de Santana), eusebio.valero@upm.es
(Eusebio Valero), esteban.ferrer@upm.es (Esteban Ferrer)
Preprint submitted to Expert Systems With Applications October 4, 2022
arXiv:2210.00101v1 [physics.flu-dyn] 30 Sep 2022
entire frequency range), while maintaining favorable aerodynamic properties
in terms of lift and drag.
Keywords: Aeroacoustics, Optimization design, Amiet theory, Machine
learning, Autoencoder
Contents
1 Introduction 3
2 Far-field noise prediction 5
2.1 The acoustic model . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Aerodynamic simulation based on XFOIL . . . . . . . . . . . 6
2.3 Methodology validation: NACA0012 airfoil . . . . . . . . . . . 7
3 Optimization framework 9
3.1 Airfoil shape parametrization . . . . . . . . . . . . . . . . . . 9
3.1.1 CST parametrization . . . . . . . . . . . . . . . . . . . 9
3.1.2 Autoencoder-based parametrization . . . . . . . . . . . 10
3.2 Variational autoencoder parameters . . . . . . . . . . . . . . . 13
3.2.1 Evaluation metric . . . . . . . . . . . . . . . . . . . . . 13
3.2.2 Loss scaling parameter k . . . . . . . . . . . . . . . . . 14
3.2.3 Autoencoder architecture . . . . . . . . . . . . . . . . . 14
3.2.4 Number of latent variables . . . . . . . . . . . . . . . . 16
3.3 Optimization algorithm . . . . . . . . . . . . . . . . . . . . . . 17
3.4 Problem formulation: optimization objective and constraints . 18
4 Results and discussion 21
4.1 Classical optimization with CST parametrization . . . . . . . 22
4.2 Improved optimization with autoencoder-based parametrization 22
5 Conclusions 27
6 Acknowledgements 29
Appendix A Parameters in Amiet’s theory 29
Appendix A.1 Spanwise correlation length . . . . . . . . . . . . 29
Appendix A.2 Wall-pressure spectrum . . . . . . . . . . . . . . 29
Appendix A.3 Boundary layer parameters . . . . . . . . . . . . 31
2
Appendix A.3.1 Boundary layer thickness . . . . . . . . . . 31
Appendix A.3.2 Mean velocity profile . . . . . . . . . . . . . 31
Appendix A.3.3 Turbulence intensities . . . . . . . . . . . . 32
Appendix A.3.4 Integral length scale . . . . . . . . . . . . . 32
Appendix A.3.5 Velocity spectrum . . . . . . . . . . . . . . 33
1. Introduction
Wind turbine noise has typically been one of the main drawbacks of the
wide deployment and acceptance of wind devices for the generation of clean
energy. The interest in controlling turbine acoustics is increasing since tur-
bines are now being integrated in urban environments, where noise nuisances
can be of great concern [1, 2]. Wind turbine aeroacoustics is a complex
phenomenon that includes several sources of noise. The most critical noise
sources of a typical wind turbine are: 1) steady loading, which relates to
the distribution of forces along the blade leading to broadband noise; 2) un-
steady loading caused by the incoming sheared and turbulent atmospheric
flow, associated with low frequency noise; and 3) airfoil self-noise, which en-
compasses various high frequency phenomena that relate to the boundary
layer and eddies generated as the air passes the blades. Airfoil self-noise
[3] is the minimum amount of noise produced by an aerodynamic surface
and is the main noise source in modern wind turbines [4, 5]. It is caused
by the interaction of the turbulent boundary layer with the blade surface
close to the trailing edge, when the hydrodynamic pressure fluctuations orig-
inated in the airfoil surface are scattered to the far field as noise, due to
the sudden change of impedance in the blade trailing edge discontinuity [6].
To generate quiet wind turbines, it is necessary to select airfoils that mini-
mize noise generation in operational conditions (e.g. attached flow). Typical
turbine airfoils were designed to be insensitive to roughness, by promoting
the transition near the leading edge [7], while noise generation was often
neglected. However, social pressure to minimize wind farm noise leads to
considering acoustics during the design phase of new airfoils. The SIROCO
project [8, 9, 10] incorporated aeroacoustic predictions for shape optimiza-
tion design. Lutz et al. [9] proposed a method to predict airfoil trailing edge
far-field noise, by combining XFOIL, a finite-difference code, and a modi-
fied TNO-TPD model. The method has been used to design new, less-noisy
airfoils without loss in aerodynamic performance. Similarly, Hao et al. [11]
introduced aerodynamic and aeroacoustic optimization of the wind turbine
3
blade using a genetic algorithm. Lee et al. [12] optimized the airfoils of the
wind turbine based on genetic algorithms, to reduce airfoil self-noise, and
validated the results by wind tunnel experiment. Rodrigues and Marta [13]
performed multi-objective optimization to design wind turbine blades, where
an increase in annual energy production of 15% was achieved with a reduction
in noise levels of 9.8%. Zhou et al. [14] proposed a discrete adjoint frame-
work for unsteady aerodynamic aeroacoustic optimization. More recently,
Volkmer and Carolus [15] performed aeroacoustic airfoil shape optimization
based on semi-empirical aeroacoustic models. There, XFOIL is combined
with Amiet’s model and evolutionary optimization to design airfoils with less
noise while maintaining the required lift. Ricks et al. [16] proposed multi-
objective aerodynamic-aeroacoustic shape optimization of airfoils based on
a Reynolds-averaged Navier-Stokes solver with a state-of-the-art wall pres-
sure spectrum model and Amiet’s model for trailing edge noise. Bu et al.
[17] developed the framework for aerodynamic/aeroacoustic variable-fidelity
optimization of helicopter rotor based on hierarchical Kriging model. These
works show the efficiency of using Amiet’s theory and XFOIL for aeroacoustic
optimization, whereas recent machine learning strategies (e.g., autoencoders)
have not been considered.
Machine learning (ML) techniques are permeating in all fields of fluid
dynamics [18, 19, 20, 21, 22] and aerodynamic optimization; see the recent
review by Li et al. [23]. In particular, unsupervised techniques for dimen-
sionality reduction enable the treatment of problems with a large number
of variables by coding the information into a reduced number of parameters
(i.e., latent variables). Neural networks are ideal for representing nonlinear
functions and can thus encode information very efficiently in very few latent
variables. Autoencoders (AEs) are artificial neural networks that can reduce
the dataset information (encoding), but that can also reconstruct (decode)
the original data, minimizing the reconstruction error. Variational autoen-
coders (VAEs) improve the behavior of classic autoencoders by enforcing
latent variables to follow smooth statistical distributions (typically normal
distributions) [24, 25]. By doing so, it is possible to avoid discontinuities in
the latent variables, which helps to reconstruct the original database. An-
other advantage of requiring smooth latent variables is that they can be used
for optimization.
VAEs have not been widely adopted for optimization. In fact, to the au-
thors’ knowledge, only Rios et al. [26] have used VAEs for car optimization.
They showed the superiority of this technique in incorporating local geomet-
4
ric features, when compared to modal techniques (e.g., principal component
analysis). Additionally, Yonekura and Suzuki [27] have shown the advantages
of VAEs to represent airfoils using latent variables, but did not perform op-
timizations.
Encouraged by the previous work, in this work we propose using VAEs to
enhance the aeroacoustic optimization of wind turbine airfoils. We target the
airfoil self-noise using Amiet’s theory and develop an optimization framework
that targets improved aerodynamics with minimum noise in the design. We
include VAEs for the parametrization of the airfoil geometry and compare
this technique with the CST method [28] to parametrize the airfoil geometry.
As shown in the results, VAEs show superiority and enhanced robustness over
CST.
The remaining of this paper is organized as follows. Section 2 introduces
the far-field noise prediction framework, including the aeroacoustic and aero-
dynamic models, as well as the validation of the framework. The optimiza-
tion framework is introduced in Section 3, where the shape parametrization
including CST and VAEs, the optimization algorithm, and the problem for-
mulation are detailed. The optimization results are summarized in Section
4. Conclusions are provided in Section 5.
2. Far-field noise prediction
This section introduces the computational models used in the present
study, including the aerodynamic and acoustic models. Amiet’s theory [29]
is adopted here for the trailing edge noise prediction. The theory calculates
the far-field acoustic pressure spectrum using the wall pressure wavenumber-
frequency spectral density in the vicinity of the trailing edge as input. This
theory assumes a large span, a stationary observer and airfoil, and a uni-
form flow, and that the boundary layer turbulence is convecting over the
trailing edge as a frozen pattern, i.e., the turbulence is not affected by the
discontinuity of the trailing edge.
2.1. The acoustic model
Equation 1 presents the far-field power spectral density of an airfoil of
chord cand span bfor an observer perpendicular to the trailing edge at
5
摘要：

AeroacousticairfoilshapeoptimizationenhancedbyautoencodersJiaqingKoua,LauraBotero-Bolvarb,RomanBallanoa,OscarMarinoa,LeandrodeSantanab,EusebioValeroa,c,EstebanFerrera,caETSIAE-UPM-SchoolofAeronautics,UniversidadPolitecnicadeMadrid,PlazaCardenalCisneros3,E-28040Madrid,SpainbDepartmentofThermalFlu...

展开>> 收起<<
Aeroacoustic airfoil shape optimization enhanced by autoencoders_2.pdf

共40页,预览5页

还剩页未读， 继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

相关推荐

更多
立即下载
分类：图书资源 价格：10玖币 属性：40 页 大小：768KB 格式：PDF 时间：2025-04-27

开通VIP享超值会员特权

  • 多端同步记录
  • 高速下载文档
  • 免费文档工具
  • 分享文档赚钱
  • 每日登录抽奖
  • 优质衍生服务

作者详情

相关内容

更多

热门标签

人际关系 动力学连接体 力的合成 配电装置 高考理综 全宋诗作者索引 公务员考试
/ 40
评分 收藏
分享
立即下载
客服
关注