平面叶栅气动噪声的高精度谱元方法求解

本文建立了一种用于气动噪声预测的高精度三角谱元方法。结合CBS法(Characteristic-based split method)求解Navier-Stokes方程获取伪声压声源后,基于波动方程求解声传播问题。谱元法的谱收敛特性满足了气动声学问题的高精度需求,而CBS法的引入保证了高雷诺数问题的计算稳定性。通过两组基准解问题求解验证了本文方法的正确性。将本文方法应用于平面叶栅气动噪声计算,并考察了不同攻角下噪声的变化规律。研究内容为进一步探究各类流体机械的气动噪声提供了一种新的途径。     

基于深度神经网络的水声信号恢复方法研究*

针对干扰或噪声环境下水声目标信号难以获取的问题,该文提出研究基于深度神经网络的自适应水声被动信号波形恢复方法。在单阵元情况下,该方法提取对数功率谱特征作为输入,采用深度神经网络回归模型自适应学习目标信号的自身特征,输出降噪后的对数功率谱特征并还原时域波形。在多阵元情况下,提出阵列深度神经网络降噪方法,将部分或全部阵元特征拼接为长向量作为输入,从而利用空域信息。为全面利用阵列丰富的时频域信息,该文提出一种两阶段特征融合深度神经网络,在第一阶段将阵列分为若干个子阵,将每个子阵分别用阵列深度神经网络进行处理,在第二阶段将第一阶段的各子阵处理结果与阵列接收信号同时输入一个深度神经网络进行融合学习。实验表明,所提出的单阵元和两阶段融合深度神经网络取得了显著优于常规波束形成的恢复结果,能够准确估计目标信号波形和功率并显著提高输出信噪比。     

A numerical study of the effects of design parameters on the acoustics noise of a high efficiency propeller

A numerical study of a high efficiency propeller in the aerodynamic noise generation is carried out. Based on RANS, three-dimensional numerical simulation is performed to obtain the aerodynamic performance of the propeller. The result of the aerodynamic analysis is given as input of the acoustic calculation. The sound is calculated using the Farassat 1A, which is derived from Ffowcs Williams–Hawkings equation, and compared with the data of wind tunnel. The propeller is modified for noise reduction by changing its geometrical parameters such as diameter, chord width and pitch angle. The trend of variation between aerodynamic analysis data and acoustic calculation result are compared and discussed for different modification tasks. Meaningful conclusions are drawn on the noise reduction of propeller.     

基于时延神经网络模型的舰船辐射噪声目标识别*

水声目标被动识别是水声信号处理领域的研究热点之一。海洋环境中存在的不规则噪声干扰,使得基于传统方法的水声目标被动识别技术在实际的应用场景中效果不佳。本文采用一种基于时延网络(Time Delay Neural Network,TDNN)模型的舰船辐射噪声目标识别方法,该方法利用目标的短时平稳特性和长时关联特性对目标的声纹特征进行建模,使用梅尔谱图提取目标信号的初级特征,再通过融合注意力机制和时延神经网络的深度学习模型实现高级特性提取,最后再利用余弦相似度实现不同目标的类别划分。该方法在ShipsEar数据集和自行采集的数据进行测试验证,目标识别准确率分别达到79.2%和73.9%,可证明本文方法的有效性。     

襟翼侧缘噪声机理及修型降噪设计

襟翼侧缘噪声是飞机起降阶段机体噪声的重要噪声源。采用极大涡模拟对襟翼侧缘非定常流场进行数值模拟,分析其噪声产生机理.基于此,提出了两种襟翼侧缘修型方式,应用虚拟渗透面的Ffowcs Williams and Hawkings(FW-H)声比拟方法将修型构型的远场噪声频谱特性和指向性与基准构型对比分析,研究其降噪效果。通过流场和声场的数值模拟表明,襟翼侧缘噪声属于宽频噪声。不同的襟翼侧缘形状改变了流场形态、侧缘涡结构以及涡系的发展过程,进而对声源分布和远场噪声特性产生影响。结果表明:在给定的5°计算迎角下,两种襟翼侧缘修型方式在保证增升装置的原有升阻气动特性的前提下,能达到减小全场总声压级1~2 dB的降噪效果。      

基于密集卷积和域自适应的高光谱图像分类

针对常规的高光谱图像分类算法不能很好地解决不同图像中的频谱偏移的问题,提出了一种基于密集卷积和域自适应的高光谱图像分类算法,首先在源域中使用密集卷积进行深度特征学习,然后应用域自适应技术转移到目标域。目前的域自适应高光谱图像分类框架中常用卷积神经网络进行特征学习,但是当深度增加时会出现因梯度消失而导致分类精度下降的情况,因此本文通过引入密集卷积进行深度特征学习,提高域自适应高光谱图像分类的精度。在Indiana高光谱数据集和Pavia高光谱数据集上验证所提算法的有效性,整体分类精度分别为61.06%和89.63%,与其他域自适应高光谱图像分类方法对比,所提方法具有更好的分类精度。     

A deep neural network surrogate modeling benchmark for temperature field prediction of heat source layout

The thermal issue is of great importance during the layout design of heat source components in systems engineering, especially for high functional-density products. Thermal analysis requires complex simulation, which leads to an unaffordable computational burden to layout optimization as it iteratively evaluates different schemes. Surrogate modeling is an effective method for alleviating computation complexity. However, the temperature field prediction(TFP) with complex heat source layout(HSL) input is an ultra-high dimensional nonlinear regression problem, which brings great difficulty to traditional regression models. The deep neural network(DNN) regression method is a feasible way for its good approximation performance. However, it faces great challenges in data preparation for sample diversity and uniformity in the layout space with physical constraints and proper DNN model selection and training for good generality, which necessitates the efforts of layout designers and DNN experts. To advance this cross-domain research, this paper proposes a DNN-based HSL-TFP surrogate modeling task benchmark. With consideration for engineering applicability, sample generation, dataset evaluation, DNN model, and surrogate performance metrics are thoroughly investigated. Experiments are conducted with ten representative state-of-the-art DNN models. A detailed discussion on baseline results is provided, and future prospects are analyzed for DNN-based HSL-TFP tasks.     

Helicopter noise in hover: Computational modelling and experimental validation

The aeroacoustic characteristics of a helicopter rotor are calculated by a new method, to assess its applicability in assessing rotor performance in hovering. Direct solution of the Euler equations in a noninertial coordinate system is used to calculate the near-field flow around the spinning rotor. The far-field noise field is calculated by the Ffowcs Williams–Hawkings (FW–H) method using permeable control surfaces that include the blade. For a multiblade rotor, the signal obtained is duplicated and shifted in phase for each successive blade. By that means, the spectral characteristics of the far-field noise may be obtained. To determine the integral aerodynamic characteristics of the rotor, software is written to calculate the thrust and torque characteristics from the near-field flow solution. The results of numerical simulation are compared with experimental acoustic and aerodynamic data for a large-scale model of a helicopter main rotor in an open test facility. Two- and four-blade configurations of the rotor are considered, in different hover conditions. The proposed method satisfactorily predicts the aerodynamic characteristics of the blades in such conditions and gives good estimates for the first harmonics of the noise. That permits the practical use of the proposed method, not only for hovering but also for forward flight.     

水下高分辨率声图中小目标的深度网络分类方法

针对声成像数据缺少条件下的水下沉底小目标分类问题,提出一种深度网络分类算法。首先,采用高斯混合模型对声影区统计特性进行建模并提取声图阴影,在此基础上构建仿真数据集和真实数据集。将仿真数据集输入卷积神经网络进行训练,保留其特征提取部分,用于对真实数据集进行特征提取.重建网络分类部分并采用真实数据集的特征向量进行训练。结果表明,所提出的方法分类正确率可达88.24%,与6种对照方法相比平均分类正确率分别提升8.67%,20.47%,19.78%,11.59%,9.01%,11.58%。验证了所提出方法在小样本条件下具有较好对水下沉底小目标的分类能力。其学习曲线收敛到96.25%,仅比验证曲线高5.14%,说明在一定程度上缓解了过拟合问题。将改进的卷积神经网络应用于融合分类器,通过与逻辑回归分类器、支持向量机对目标进行分类并融合决策,正确率为93.33%,可进一步提高算法的正确率和稳定性.      

用于自适应波束形成语音增强的球谐域掩蔽函数估计方法

提出一种用于球形阵列自适应波束形成的掩蔽函数估计方法。该方法利用包含空间信息的球谐系数提取低维空间向量,并采用复高斯混合模型和深度学习两种方案来估计掩蔽函数,最终利用估计的掩蔽函数设计最小方差无失真响应波束形成器,以达到空域滤波的效果。理论分析和仿真实验证明,对于相同时长的声信号,球谐域掩蔽函数估计方法的计算复杂度比传统阵元域估计方法低了一个数量级。并且在大部分声场环境中,尤其在低信噪比情况下,所提方法的语音质量感知评估测度得分、分段信噪比和短时客观可懂度明显高于阵元域方法,三者最高分别可提升1.31 dB,4.54 dB和35%。另外,实际声学环境的测量实验也验证了所提方法在不影响可懂度的条件下比传统阵元域方法具备更高的降噪量。      

Extraction of the acoustic component of a turbulent flow exciting a plate by inverting the vibration problem

An improvement of the Force Analysis Technique (FAT), an inverse method of vibration, is proposed to identify the low wavenumbers including the acoustic component of a turbulent flow that excites a plate. This method is a significant progress since the usual techniques of measurements with flush-mounted sensors are not able to separate the acoustic and the aerodynamic energies of the excitation because the aerodynamic component is too high. Moreover, the main cause of vibration or acoustic radiation of the structure might be due to the acoustic part by a phenomenon of spatial coincidence between the acoustic wavelengths and those of the plate. This underlines the need to extract the acoustic part. In this work, numerical experiments are performed to solve both the direct and inverse problems of vibration. The excitation is a turbulent boundary layer and combines the pressure field of the Corcos model and a diffuse acoustic field. These pressures are obtained by a synthesis method based on the Cholesky decomposition of the cross-spectra matrices and are used to excite a plate. Thus, the application of the inverse problem FAT that requires only the vibration data shows that the method is able to identify and to isolate the acoustic part of the excitation. Indeed, the discretization of the inverse operator (motion equation of the plate) acts as a low-pass wavenumber filter. In addition, this method is simple to implement because it can be applied locally (no need to know the boundary conditions), and measurements can be carried out on the opposite side of the plate without affecting the flow. Finally, an improvement of FAT is proposed. It regularizes optimally and automatically the inverse problem by analyzing the mean quadratic pressure of the reconstructed force distribution. This optimized FAT, in the case of the turbulent flow, has the advantage of measuring the acoustic component up to higher frequencies even in the presence of noise.     

高速动车组气动噪声试验与仿真分析*

通过风洞试验对某高速动车组整车、受电弓及转向架远场气动噪声特性进行分析。试验结果表明,高速动车组远场气动噪声是一宽频噪声,总声能随速度的6.6次方增加;由受电弓引起的远场气动噪声主要集中在中高频,噪声峰值频率随速度变化线性增加;由转向架引起的远场气动噪声主要集中在中低频,噪声峰值频率与速度无关。在此基础上,通过大涡模拟和声扰动方程获得该高速动车组近场噪声。高速动车组远场噪声测点仿真结果与试验结果的最大差值2.2 dB(A),最大相对误差2.5%,表明仿真模型的准确性。仿真结果表明,车头近场噪声以车头鼻尖为界,底部气动噪声能量大于上部流线型气动噪声能量,其中转向架舱位置噪声能量最大,因此进行车内外降噪方案设计时,应重点关注车头转向架舱位置。     

Machine Learning and Micromagnetic Studies of Magnetization Switching

Magnetization switching is one of the most fundamental topics in the field of magnetism.Machine learning(ML)models of random forest(RF),support vector machine(SVM),deep neural network(DNN)methods are built and trained to classify the magnetization reversal and non-reversal cases of single-domain particle,and the classification performances are evaluated by comparison with micromagnetic simulations.The results show that the ML models have achieved great accuracy and the DNN model reaches the best area under curve(AUC)of 0.997,even with a small training dataset,and RF and SVM models have lower AUCs of 0.964 and 0.836,respectively.This work validates the potential of ML applications in studies of magnetization switching and provides the benchmark for further ML studies in magnetization switching.     

应用深度学习识别法兰螺栓连接状态

针对常规诊断方法对螺栓的连接状态识别效果差、鲁棒性和抗噪性弱等问题,提出了基于深度学习理论的螺栓检测新方法。首先以4种预紧力状态下的法兰螺栓结构产生的声发射信号为研究对象,借助于自适应噪声的完整集成经验模态分解理论以及梅尔频率倒谱系数特征提取方式,实现了声发射信号的自适应消噪和最优模态函数分量组的选取,提取到了可以较好分辨螺栓连接状态的梅尔频率倒谱系数特征值。通过训练模型,较好地对4种连接状态下的螺栓进行了识别。结果表明,该模型在法兰螺栓的声发射信号的诊断中,准确率高,具有较好的抗噪性和鲁棒性。     

相干声线跟踪理论中的周期界面迭代散射 仿真方法*

界面声反射模拟是室内复杂声学现象仿真的关键。针对传统声学仿真方法对于周期散射结构存在条件下声场仿真精度较低的问题,本文发展了一种基于迭代散射模型的室内相干声线跟踪法。此方法以经典的相干声线跟踪法为基础,将室内中常见的周期散射结构进行几何形状上的简化处理,然后依据周期散射定理给出声波在界面上的散射方向及能量,并将原始声线迭代分裂为相应的散射子声线,继续对其跟踪处理,此迭代散射模型对周期散射结构上的界面散射现象进行了准确的模拟。数值验证结果表明,本文方法可以有效地在低频段提高室内声场仿真精度,可为具有复杂散射现象的室内仿真提供新思路。     

Handwritten digit recognition based on ghost imaging with deep learning

We present a ghost handwritten digit recognition method for the unknown handwritten digits based on ghost imaging(GI) with deep neural network, where a few detection signals from the bucket detector, generated by the cosine transform speckle, are used as the characteristic information and the input of the designed deep neural network(DNN), and the output of the DNN is the classification. The results show that the proposed scheme has a higher recognition accuracy(as high as98% for the simulations, and 91% for the experiments) with a smaller sampling ratio(say 12.76%). With the increase of the sampling ratio, the recognition accuracy is enhanced. Compared with the traditional recognition scheme using the same DNN structure, the proposed scheme has slightly better performance with a lower complexity and non-locality property.The proposed scheme provides a promising way for remote sensing.     

基于深度张量神经网络预测有机分子的相互作用能

分子的相互作用在分子动力学模拟过程中起着关键的作用. 受限于计算资源,大分子的长时间尺度的相互作用能无法通过量化计算实现. 本文采用一种深度学习框架-深度张量神经网络来预测三个有机分子相关体系中量化精度的相互作用能. 其中,分子的几何结构和原子类型作为网络的输入用于预测相互作用能. 通过分层生成的数据集合实现了网络中隐层参数的优化和训练. 相互作用能的预测结果显示,深度张量神经网络可以在较短的时间内,在1 kcal/mol的平均绝对误差的范围内准确预测分子间的相互作用能. 这一过程提高了计算效率,并为计算相互作用能提供了可靠的计算框架.     

基于蜗舌改型的空调用离心风机流动分析及降噪研究

本文采用计算流体动力学和声类比相结合的混合方法对空调用离心风机进行流场以及声场的计算,同时进行风机风量和噪声的实验测量,验证所采用的数值计算模型和计算方法的有效性.针对原型非常规蜗壳,提取蜗壳中间截面型线进行直蜗舌的蜗壳设计,在此基础上设计了三种倾斜蜗舌的蜗壳.根据数值计算结果,对最优倾斜蜗舌进行了实验验证。经实验测试,风机在各个工况点风量均有提升,在最大风量点风量提升6.0%,噪声降低1.4 dB(A).数值分析风机内部流动特征及噪声特性,发现在蜗舌附近流动区域内湍流强度和涡量明显减小,在叶片通过频率处声功率谱密度以及噪声峰值明显下降,这也表明风机的旋转噪声得到了有效控制。     

Optimal design of acoustic performance for automotive air-cleaner

The research on optimal design of acoustic performance for air-cleaner was presented and evaluated in this paper using a numerical simulation method and experimental method. A modified measurement was proposed in the paper to solve the problem of air flow noise caused by the original measurement layout. The modified measurement was proved to produce good results. A new study on the acoustic effect of filter was also developed here and noise reduction (NR) of the air-cleaner was predicted based on the study using the numerical simulation method. Actual intake noise was then predicted using the simulation method. Side branch silencers were designed to improve the acoustic capacity of the air-cleaner. The result shows that the newly optimized air-cleaner has a much better performance of noise elimination when engine operating.     

Deep learning-based energy-efficient relay precoder design in MIMO-CRNs

This paper investigates the problem of energy efficient relay precoder design in multiple-input multiple-output cognitive relay networks (MIMO-CRNs). This is a non-convex fractional programming problem, which is traditionally solved using computationally expensive optimization methods. In this paper, we propose a deep learning (DL) based approach to compute an approximate solution. Specifically, a deep neural network (DNN) is employed and trained using offline computed optimal solution. The proposed scheme consists of an offline data generation phase, an offline training phase, and an online deployment phase. The numerical results show that the proposed DNN provides comparable performance at significantly lower computational complexity as compared to the conventional optimization-based algorithm that makes the proposed approach suitable for real-time implementation.