基于三维N型卷积神经网络和频域注意力–亥姆霍兹正则化的近场声源重建方法

籍宇阳; 王德禹

doi:10.19693/j.issn.1673-3185.03127

基于三维N型卷积神经网络和频域注意力–亥姆霍兹正则化的近场声源重建方法

doi: 10.19693/j.issn.1673-3185.03127

籍宇阳^{1, 2,},
王德禹^{1, 2, ,}

1.
上海交通大学海洋工程国家重点实验室，上海 200240
2.
上海交通大学海洋装备研究院，上海 200240

详细信息

作者简介:
籍宇阳，男，1998年生，硕士生。研究方向：舰船噪声源识别。Email：jiyuyang@sjtu.edu.cn

王德禹，男，1963年生，博士，教授，博士生导师。研究方向：船舶与海洋工程结构力学，结构优化设计与可靠性分析，结构极限强度与试验技术研究。E-mail：dywang@sjtu.edu.cn

通信作者:
王德禹

中图分类号: TB52；U661.44
计量
- 文章访问数: 163
- HTML全文浏览量: 23
- PDF下载量: 2
- 被引次数: 0
出版历程
- 收稿日期: 2022-10-12
- 修回日期: 2022-12-22
- 网络出版日期: 2022-12-26

Near-field acoustic reconstruction method based on three-dimensional N-shaped convolution neural network and frequency focal-KH regularization

JI Yuyang^{1, 2
,},
WANG Deyu^{1, 2
, ,}

1.
State Key Laboratory of Ocean Engineering, Shanghai Jiao Tong University, Shanghai 200240, China
2.
Institute of Marine Equipment, Shanghai Jiao Tong University, Shanghai 200240, China

基于三维N型卷积神经网络和频域注意力–亥姆霍兹正则化的近场声源重建方法由籍宇阳，等创作，采用知识共享署名4.0国际许可协议进行许可。

摘要

摘要: 目的针对全息面、低采样率条件下近场声源重建误差较大的问题，提出一种高分辨率、低误差的平面声源表面法向振速重建的深度神经网络框架。方法首先，建立用于近场声源重建问题的三维N型卷积神经网络框架（包含预编码器），通过提取空间声场频域内的特征，以弥补空间信息的稀疏性；然后，提出频域注意力机制，设计包含频域注意力–归一化重建均方误差、亥姆霍兹正则项的损失函数，以自适应增加频域内难训练样本的损失权重，从而提升声源在高频和本征特征区间的重建精度；最后，通过Matlab对COMSOL Multiphysics软件进行二次开发，建立矩形薄板声振模型的训练集和测试集，开展对比验证。结果对比结果表明，该方法在验证集上100～2 000 Hz范围内的平均重建误差仅为4.96%，重建精度明显高于SRCNN和PV-NN。结论该研究成果可以降低近场声源重建实船应用中的全息面采样点数量，同时可保证较高的声源面法向振速重建精度。
- 近场声源重建 /
- 声源识别 /
- 三维卷积 /
- 亥姆霍兹正则化
Abstract: Objectives Low sampling rates on reconstruction surfaces cause high reconstruction error in near-field acoustic holography. Therefore, a deep learning-based approach which is applicable to planar sound sources and high-precision reconstruction with low sampling rates is put forward. Methods A three-dimensional N-shaped convolution neural network for near-field acoustic reconstruction is established to extract features in the frequency dimension in order to make up for sparse sampling in the spatial dimension. A frequency focal mechanism, namely an adaptive frequency weight focus mechanism, is put forward to improve reconstruction precision in the natural frequency and high frequency. Moreover, this paper also raises frequency-scaled focal loss and frequency-scaled focal Kirchhoff-Helmholtz loss, which are considered regularization. To validate the proposed methods, datasets are created with COMSOL Multiphysics and Matlab. Results The mean error range of 100–2 000 Hz of the algorithm proposed in this paper is only 4.96%, higher than those of SRCNN and PV-NN. Conclusions The proposed method is verified as having the potential to reconstruct the accurate velocity fields of sound sources under low sampling rates.
- near-field acoustic reconstruction /
- sound source recognition /
- 3D convolution /
- Kirchhoff-Helmholtz regularization

HTML全文

图 1 振动声源向外辐射声场的示意图

Figure 1. Schematic diagram of vibrational sound source acoustic radiation

下载: 全尺寸图片幻灯片

图 2 声源面和全息面示意图

Figure 2. Illustration of source surface and holography surface

下载: 全尺寸图片幻灯片

图 3 深度学习训练的流程图

Figure 3. Procedure of deep learning training

下载: 全尺寸图片幻灯片

图 4 3D NCNN-NAH网络架构图

Figure 4. Framework of 3D NCNN-NAH

下载: 全尺寸图片幻灯片

图 5 预编码器中双线性插值上采样模块

Figure 5. Bilinear interpolation block in pre-encoder

下载: 全尺寸图片幻灯片

图 6 残差结构示意图

Figure 6. Illustration of residual block

下载: 全尺寸图片幻灯片

图 7 反卷积模块示意图

Figure 7. Illustration of deconvolution block

下载: 全尺寸图片幻灯片

图 8 数据集中重建声源表面重建点的法相位移随频域变化图

Figure 8. Displacement of the reconstructed points on sound source varies in frequency domain in training dataset

下载: 全尺寸图片幻灯片

图 9 不同频率下3D NCNN-NAH重建效果和仿真得到的真实值对比

Figure 9. Comparison of theoretical velocity and reconstruction effect from 3D NCNN-NAH in different frequency

下载: 全尺寸图片幻灯片

图 10 3D NCNN-NAH, SRCNN，PV-NN在不同频域范围内的误差对比

Figure 10. Comparison of the reconstruction errors of 3D NCNN, SRCNN, PV-NN in frequency domains

下载: 全尺寸图片幻灯片

图 11 在频域范围内是否包含预编码器的重建误差对比

Figure 11. Comparison of reconstruction errors with and without pre-encoder in the frequency domain

下载: 全尺寸图片幻灯片

图 12 NCNN-NAH在FSFLos，FFS-KHLoss组合监督与MSELoss监督下的频域范围内的重建误差对比

Figure 12. Comparison of reconstruction errors of NCNN-NAH in frequency domain under FSFLos, FFS-KHLoss combination supervision, and MSELoss supervision

下载: 全尺寸图片幻灯片

图 13 全息面采样点数量与重建误差的关系

Figure 13. The relationship between different sampling numbers on holography surface and reconstruction error

下载: 全尺寸图片幻灯片

图 14 不同信噪比与重建误差的关系

Figure 14. The relationship between different SNRs and reconstruction errors

下载: 全尺寸图片幻灯片

参考文献(21)

[1]	赵欣阳, 祝熠, 梅志远, 等. 不同筋材对复合材料加筋板水中透声性能影响试验研究[J]. 中国舰船研究, 2023, 18(3): 197–204. doi: 10.19693/j.issn.1673-3185.02535 ZHAO X Y, ZHU Y, MEI Z Y, et al. Experimental study on effect of different reinforcements on sound tranmission performance of composite stiffened plates in water[J]. Chinese Journal of Ship Research, 2023, 18(3): 197–204 (in Chinese). doi: 10.19693/j.issn.1673-3185.02535
[2]	李广生, 陈美霞, 原春晖. 基于陆上振动测试的水中圆柱壳结构声振响应计算方法[J]. 中国舰船研究, 2022, 17(6): 252–260. doi: 10.19693/j.issn.1673-3185.02414 LI G S, CHEN M X, YUAN C H. Calculation method of underwater acoustic and vibration response of cabin segment based on onshore vibration test[J]. Chinese Journal of Ship Research, 2022, 17(6): 252–260 (in Chinese). doi: 10.19693/j.issn.1673-3185.02414
[3]	廖健, 何琳, 陈宗斌, 等. 潜艇操舵系统噪声综述[J]. 中国舰船研究, 2022, 17(5): 74–84. doi: 10.19693/j.issn.1673-3185.02391 LIAO J, HE L, CHEN Z B, et al. Overview of submarine steering system noise[J]. Chinese Journal of Ship Research, 2022, 17(5): 74–84 (in Chinese). doi: 10.19693/j.issn.1673-3185.02391
[4]	左翔, 陈欢. 基于矢量声压组合基阵的柱面分布噪声源近场高分辨定位方法[J]. 中国舰船研究, 2017, 12(4): 147–150. doi: 10.3969/j.issn.1673-3185.2017.04.023 ZUO X, CHEN H. Near-field and high-resolution cylind-rical noise source location method based on vector sound pressure array[J]. Chinese Journal of Ship Research, 2017, 12(4): 147–150 (in Chinese). doi: 10.3969/j.issn.1673-3185.2017.04.023
[5]	胡清扬, 李灿灿, 郭世旭. 虚拟声源定位的等效源近场声全息算法[J]. 舰船科学技术, 2022, 44(11): 164–168. doi: 10.3404/j.issn.1672-7649.2022.11.034 HU Q Y, LI C C, GUO S X. Equivalent source near-field acoustic holography algorithm based on virtual sound source localization[J]. Ship Science and Technology, 2022, 44(11): 164–168 (in Chinese). doi: 10.3404/j.issn.1672-7649.2022.11.034
[6]	李彪, 李希友, 王志强, 等. 统计最优柱面近场声全息参数选取方法研究[J]. 舰船科学技术, 2018, 40(3): 120–127. LI B, LI X Y, WANG Z Q, et al. Research on parameter selection for the statistically optimal cylindrical near-field acoustical holography[J]. Ship Science and Technology, 2018, 40(3): 120–127 (in Chinese).
[7]	陈汉涛, 郭文勇, 韩江桂, 等. 船舶机舱内高频弱声源近场声全息方法[J]. 舰船科学技术, 2019, 41(11): 138–143,147. CHEN H T, GUO W Y, HAN J G, et al. Near-field acoustic holography method for high frequency weak sound source in ship cabin[J]. Ship Science and Technology, 2019, 41(11): 138–143,147 (in Chinese).
[8]	WANG J Z, ZHANG Z F, HUANG Y Z, et al. A 3D convolutional neural network based near-field acoustical holography method with sparse sampling rate on measur-ing surface[J]. Measurement, 2021, 177: 109297. doi: 10.1016/j.measurement.2021.109297
[9]	WILLIAMS E G. Fourier acoustics: sound radiation and nearfield acoustical holography[M]. San Diego: Academic Press, 1999: 67-89.
[10]	CHARDON G, DAUDET L, PEILLOT A, et al. Near-field acoustic holography using sparse regularization and compressive sampling principles[J]. The Journal of the Acoustical Society of America, 2012, 132(3): 1521–1534. doi: 10.1121/1.4740476
[11]	FERNANDEZ-GRAND E, XENAKI A. Compressive sensing with a spherical microphone array[J]. The Journal of the Acoustical Society of America, 2016, 139(2): EL45–EL49. doi: 10.1121/1.4942546
[12]	HALD J. A comparison of iterative sparse equivalent source methods for near-field acoustical holography[J]. The Journal of the Acoustical Society of America, 2018, 143(6): 3758–3769. doi: 10.1121/1.5042223
[13]	HALD J. Fast wideband acoustical holography[J]. The Journal of the Acoustical Society of America, 2016, 139(4): 1508–1517. doi: 10.1121/1.4944757
[14]	FERNANDEZ-GRAND E, XENAKI A, GERSTOFT P. A sparse equivalent source method for near-field acoustic holography[J]. The Journal of the Acoustical Society of America, 2017, 141(1): 532–542. doi: 10.1121/1.4974047
[15]	BI C X, LIU Y, XU, L, et al. Sound field reconstruction using compressed modal equivalent point source method[J]. The Journal of the Acoustical Society of America, 2017, 141(1): 73–79. doi: 10.1121/1.4973567
[16]	BI C X, ZHANG F M, ZHANG X Z, et al. Sound field reconstruction using block sparse Bayesian learning equivalent source method[J]. The Journal of the Acoustical Society of America, 2022, 151(4): 2378–2390. doi: 10.1121/10.0010103
[17]	伍松, 魏晟弘, 吴小龙. 压缩感知等效源法对板件近场重构精度改进的研究[J]. 机械科学与技术, 2023, 42(6): 870-877. WU S, WEI S H, WU X L, Improvement of near-field reconstruction accuracy of plate using compressed sensing equivalent source method[J]. Mechanical Science and Technology for Aerospace Engineering, 2023, 42(6): 870-877 (in Chinese).
[18]	JANSSENS O, SLAVKOVIKJ V, VERVISCH B, et al. Convolutional neural network based fault detection for rotating machinery[J]. Journal of Sound and Vibration, 2016, 377: 331–345. doi: 10.1016/j.jsv.2016.05.027
[19]	ZHANG Y Y, LI X Y, GAO L, et al. Intelligent fault diagnosis of rotating machinery using a new ensemble deep auto-encoder method[J]. Measurement, 2020, 151: 107232. doi: 10.1016/j.measurement.2019.107232
[20]	OLIVIERI M, PEZZOLI M, MALVERMI R, et al. Near-field acoustic holography analysis with convolutional neural networks[C]//INTER-NOISE and NOISE-CON Congress and Conference Proceedings. Seoul, Korea: Institute of Noise Control Engineering, 2020: 5607−5618.
[21]	OLIVIERI M, PEZZOLI M, ANTONACCI F, et al. A physics-informed neural network approach for nearfield acoustic holography[J]. Sensors, 2021, 21(23): 7834. doi: 10.3390/s21237834