Time-domain dynamic constitutive model suitable for mucky soil site seismic response

Soil nonlinear behavior displays noticeable effects on the site seismic response.This study proposes a new functional expression of the skeleton curve to replace the hyperbolic skeleton curve.By integrating shear modulus and combining the dynamic skeleton curve and the damping degradation coefficient,the constitutive equation of the logarithmic dynamic skeleton can be obtained,which considers the damping effect in a soil dynamics problem.Based on the finite difference method and the multi-transmitting boundary condition,a 1D site seismic response analysis program called Soilresp1D has been developed herein and used to analyze the time-domain seismic response in three types of sites.At the same time,this study also provides numerical simulation results based on the hyperbolic constitutive model and the equivalent linear method.The results verify the rationality of the new soil dynamic constitutive model.It can analyze the mucky soil site nonlinear seismic response,reflecting the deformation characteristics and damping effect of the silty soil.The hysteresis loop area is more extensive,and the residual strain is evident.

Time-Domain Analysis of Body Freedom Flutter Based on 6DOF Equation

The reduced weight and improved efficiency of modern aeronautical structures result in a decreasing separation of frequency ranges of rigid and elastic modes.Particularly,a high-aspect-ratio flexible flying wing is prone to body freedomflutter(BFF),which is a result of coupling of the rigid body short-periodmodewith 1st wing bendingmode.Accurate prediction of the BFF characteristics is helpful to reflect the attitude changes of the vehicle intuitively and design the active flutter suppression control law.Instead of using the rigid body mode,this work simulates the rigid bodymotion of the model by using the six-degree-of-freedom(6DOF)equation.A dynamicmesh generation strategy particularly suitable for BFF simulation of free flying aircraft is developed.An accurate Computational Fluid Dynamics/Computational Structural Dynamics/six-degree-of-freedom equation(CFD/CSD/6DOF)-based BFF prediction method is proposed.Firstly,the time-domain CFD/CSD method is used to calculate the static equilibrium state of the model.Based on this state,the CFD/CSD/6DOF equation is solved in time domain to evaluate the structural response of themodel.Then combinedwith the variable stiffnessmethod,the critical flutter point of the model is obtained.This method is applied to the BFF calculation of a flyingwing model.The calculation results of the BFF characteristics of the model agree well with those fromthe modalmethod andNastran software.Finally,the method is used to analyze the influence factors of BFF.The analysis results show that the flutter speed can be improved by either releasing plunge constraint or moving the center ofmass forward or increasing the pitch inertia.

Health Monitoring of Milling Tool Inserts Using CNN Architectures Trained by Vibration Spectrograms

In-process damage to a cutting tool degrades the surface􀀀nish of the job shaped by machining and causes a signi􀀀cant􀀀nancial loss.This stimulates the need for Tool Condition Monitoring(TCM)to assist detection of failure before it extends to the worse phase.Machine Learning(ML)based TCM has been extensively explored in the last decade.However,most of the research is now directed toward Deep Learning(DL).The“Deep”formulation,hierarchical compositionality,distributed representation and end-to-end learning of Neural Nets need to be explored to create a generalized TCM framework to perform eciently in a high-noise environment of cross-domain machining.With this motivation,the design of dierent CNN(Convolutional Neural Network)architectures such as AlexNet,ResNet-50,LeNet-5,and VGG-16 is presented in this paper.Real-time spindle vibrations corresponding to healthy and various faulty con􀀀gurations of milling cutter were acquired.This data was transformed into the time-frequency domain and further processed by proposed architectures in graphical form,i.e.,spectrogram.The model is trained,tested,and validated considering dierent datasets and showcased promising results.

An Improved Time-Domain Inverse Technique for Localization and Quantifcation of Rotating Sound Sources

The time-domain inverse technique based on the time-domain rotating equivalent source method has been proposed to localize and quantify rotating sound sources. However, this technique encounters two problems to be addressed: one is the time-consuming process of solving the transcendental equation at each time step, and the other is the difculty of controlling the instability problem due to the time-varying transfer matrix. In view of that, an improved technique is proposed in this paper to resolve these two problems. In the improved technique, a de-Dopplerization method in the time-domain rotating reference frame is frst applied to eliminate the Doppler efect caused by the source rotation in the measured pressure signals, and then the restored pressure signals without the Doppler efect are used as the inputs of the time-domain stationary equivalent source method to locate and quantify sound sources. Compared with the original technique, the improved technique can avoid solving the transcendental equation at each time step, and facilitate the treatment of the instability problem because the transfer matrix does not change with time. Numerical simulation and experimental results show that the improved technique can eliminate the Doppler efect efectively, and then localize and quantify the rotating nonstationary or broadband sources accurately. The results also demonstrate that the improved technique can guarantee a more stable reconstruction and compute more efciently than the original one.

Integrated system of traditional THz time-domain spectroscopy and asynchronous optical sampling

Terahertz time-domain spectroscopy(THz-TDS)system,as a new means of spectral analysis and detection,plays an increasingly pivotal role in basic scientific research.However,owing to the long scanning time of the traditional THz-TDS system and the complex control of the asynchronous optical scanning(ASOPS)system,which requires frequent calibration,we combine traditional THz-TDS and ASOPS systems to form a composite system and propose an all-fiber trigger signal generation method based on the time overlapping interference signal generated by the collinear motion of two laser pulses.Finally,the time-domain and frequency-domain spectra are obtained by using two independent systems in the integrated systems.It is found that the full width at half maximum(FWHM)of the time-domain spectra and the spectral width of the frequency-domain spectra are almost the same,but the sampling speed of the ASOPS system is significantly faster than that of the traditional THz-TDS system,which conduces to the study of the transient characteristics of substances.

一个面向短波通信的LHOG话音检测方法

噪声环境下语音检测准确率偏低是短波通话面临的公开挑战。当前已有方法应用有限,其根源在于难以可靠地在噪音环境下提取准确且高效的语音特征。针对上述问题,提出了一个面向短波通信的低秩方向梯度直方图(Low-rank Histogram of Oriented Gradient,LHOG)话音检测方法。首先,对目标音频源数据进行预处理,实现噪声环境下语音信息的可视化表征;然后,在HOG特征提取器中嵌入低秩化结构,缓解特征中的冗余信息,并降低噪声干扰,从而获得准确且高效的特征;最后,通过常用的SVM分类模型便可在噪声环境中准确快速地区分话音和噪声。测试结果表明,该方法的准确率达到了95.12%,误报率仅为0.96%,漏报率为13.14%。与现有主流方法的对比实验证明,该方法话音检测准确率高,资源占用少,能够有效提高短波通信侦控效率。

心音频谱图在二尖瓣反流患者左心室功能监测中的作用

目的探求使用可穿戴设备获取心音频谱图,观察不同程度二尖瓣反流(MR)患者心音频谱特征在左心室功能监测中的作用。方法入选2020年6月至2021年5月因心血管疾病入住上海交通大学医学院附属瑞金医院的102例MR患者,根据心脏超声MR反流束面积/左心房面积比值超过20%将其分为轻度MR组72例和中度及以上MR组30例。基于心音频谱图对心音持续时间、最大声强、最大心音振幅进行分析,并与心脏超声测定的左心室射血分数(LVEF)进行相关性分析。结果心音频谱图第一心音持续时间与收缩期持续时间比值(S11-S12/S11-S21)与LVEF呈线性负相关(r=-0.535,P<0.01),S11-S12/S11-S21用于判定LVEF降低(LVEF<50%)曲线下面积为0.840(95%CI=0.756~0.924,P<0.01),截断值是55.50%,敏感度78.6%,特异度83.3%。结论心音频谱图提示,MR患者心脏收缩期第一心音持续时间占比与LVEF呈负相关性,在二尖瓣病患的心功能居家监测中存在潜在的应用价值。

基于语谱图的管制员疲劳状态检测研究

现阶段利用陆空通话语音对管制员疲劳状态的研究中,大多只考虑了语音在时域或频域的变化,而忽视了疲劳会同时在时域与频域上产生影响。将三种疲劳状态下的陆空通话语音分别转化为可同时反映时域与频域特性的语音频谱图像,利用灰度共生矩阵提取四维典型的特征参数,对比管制员在不同状态下特征参数的变化情况,构建管制员疲劳检测模型并对输入特征进行检测。结果表明:利用语谱图特征结合传统特征作为输入特征的检测准确率最高,达到95.49%,较单一使用传统特征的检测准确率高出4%;管制员疲劳状态的变化会直观地反映在语谱图上,会对其特征值产生影响,利用这种影响对管制员疲劳状态进行检测,可以得到良好的检测结果。

双解码卷积循环网络风噪声有源控制

本文提出一种利用双解码卷积循环网络(Dual-decoder Convolutional Recurrent Network,DCRN)代替FxLMS(Filtered-x Least Mean Square)算法的有源噪声控制方法,考虑到相位信息在有源噪声控制(Active Noise Control,ANC)中的重要性,DCRN网络的输入特征为噪声信号的复数频谱(包括实部谱和虚部谱).网络结构中,采用编码模块从噪声复数频谱中提取特征,利用双解码模块分别估计网络输出的实部谱和虚部谱,采用参数共享机制和组策略以降低训练参数的数量并提高网络的学习能力和泛化能力.特别是针对风噪声,选用新的损失函数以及对训练数据进行正则化处理以提升DCRN的性能.实验结果表明,DCRN方法在仿真环境与有源降噪耳机环境下对一般噪声和风噪声都表现出良好的降噪性能和鲁棒性.

基于最小二乘图像坐标修正的中阶梯光谱仪谱图还原算法

中阶梯光谱仪具有交叉色散特征,二维谱图还原是决定其波长测量准确性的关键环节,但是环境变化、加工装调等引起的光斑坐标变化对谱图还原精度造成严重影响。本文提出了一种基于最小二乘图像坐标修正的中阶梯光谱仪谱图还原算法,首先提取校准汞灯光源多波长光斑质心坐标,利用理论像点和实际像点坐标构建系数矩阵,通过最小二乘法估计获得二维像面的平移、缩放、旋转系数,再采用多项式拟合减小残差影响,实现不同波长光斑的图像坐标修正,进而实现波长精确解算。实验结果显示,该算法能有效提高中阶梯光谱仪的谱图还原精度,在模拟较大装调误差条件下,修正后坐标与理想坐标偏差小于0.6个像元,证明了该算法具有较高的精度。

PC-based artif icial neural network inversion for airborne time-domain electromagnetic data

Traditionally, airborne time-domain electromagnetic （ATEM） data are inverted to derive the earth model by iteration. However, the data are often highly correlated among channels and consequently cause ill-posed and over-determined problems in the inversion. The correlation complicates the mapping relation between the ATEM data and the earth parameters and thus increases the inversion complexity. To obviate this, we adopt principal component analysis to transform ATEM data into orthogonal principal components （PCs） to reduce the correlations and the data dimensionality and simultaneously suppress the unrelated noise. In this paper, we use an artificial neural network （ANN） to approach the PCs mapping relation with the earth model parameters, avoiding the calculation of Jacobian derivatives. The PC-based ANN algorithm is applied to synthetic data for layered models compared with data-based ANN for airborne time-domain electromagnetic inversion. The results demonstrate the PC-based ANN advantages of simpler network structure, less training steps, and better inversion results over data-based ANN, especially for contaminated data. Furthermore, the PC-based ANN algorithm effectiveness is examined by the inversion of the pseudo 2D model and comparison with data-based ANN and Zhody＇s methods. The results indicate that PC-based ANN inversion can achieve a better agreement with the true model and also proved that PC-based ANN is feasible to invert large ATEM datasets.

合成语声的声学分析及识别特征算法

当前社会新型犯罪中电信诈骗案件频发,急需一种能够自动有效区分语声真伪的方法。为进一步增强目前深度学习领域识别合成语声的能力,为保障语声信息安全提供技术上的支持,针对合成语声声学特性上异于真实语声的特点,分析对比合成语声和真实语声的声学特性,设计了一种声学特征均方根角量化语声声强变化程度,结合基频变化率和语声窄带频谱图声学特征进行融合,量化了声学特性差异,聚焦了合成语声中关键声学信息。在神经网络模型中融合输入声学特征,在FoR数据集的验证集上得到了0.6%的等错误率,在测试集上最好结果达到了10.8%的等错误率。该文成功实现了对合成语声的识别,证实了声学特征的有效性和研究方案的可行性,在一定程度上拓宽了合成语声特征设计的研究思路。

基于Mel声谱图与改进SEResNet的鱼类行为识别

养殖环境中饲料投放、水流变化等刺激源导致鱼类声音分辨难,使行为识别准确率不高,为解决上述问题,提出基于Mel声谱图(Mel spectrogram)与改进SEResNet的鱼类行为识别模型TAP-SEResNet。首先针对鱼类行为声音频率波动大、特征差异小,造成特征提取难的问题,采用高分辨率、特征表示较好的Mel声谱图以捕捉鱼类声音的频谱特征。其次针对鱼类声音特征关键信息易丢失的难题,提出在SEResNet模型中融合时序聚合池化层(Temporal Aggregated Pooling,TAP),提取池化区域的最大值和平均值,保留鱼类行为更多细粒度声音特征,提高识别准确率。为验证所提模型的有效性,分别设计了消融试验和模型性能对比试验,试验结果显示:TAP-SEResNet相比SEResNet在不降低检测速度的条件下准确率提升了3.23%;相比PANNS-CNN14、ECAPA-TDNN及MFCC+ResNet等先进声音识别模型,TAP-SEResNet在准确率上分别提升了5.32%、2.80%和1.64%。所提模型有助于养殖过程中对鱼类行为实现精准监测,对精准养殖具有重要的推动作用。

Solution and Analysis of Chatter Stability for End Milling in the Time-domain

In this paper, the instantaneous undeformed chip thickness is modeled to include the dynamic modulation caused by the tool vibration while the dynamic regenerative effects are taken into account. The numerical method is used to solve the differential equations goveming the dynamics of the milling system. Several chatter detection criteria are applied synthetically to the simulated signals and the stability diagram is obtained in time-domain. The simulation results in time-domain show a good agreement with the analytical prediction, which is validated by the cutting experiments. By simulating the chatter stability lobes in the time-domain and analyzing the influences of different spindle speeds on the vibration amplitudes of the tool under a Fixed chip-load condition, conclusions could be drawn as follows： In rough milling, higher machining efficiency can be achieved by selecting a spindle speed corresponding to the axial depth of cut in accordance with the simulated chatter stability lobes, and in Fmish milling, lower surface roughness can be achieved by selecting a spindle speed well beyond the resonant frequency of machining system.

基于实体GIS设备试验的局部放电光学信号传播特性研究

光测法具有电磁免疫性,特别适用于具有封闭结构的气体绝缘开关设备(gas insulated switchgear, GIS)局部放电检测,但是光信号在传播过程中易因传播距离增加或部件遮挡而快速衰减。为此,基于110 kV实际GIS设备开展局部放电光学信号传播特性试验,研究传播距离和绝缘子遮挡对导杆尖刺缺陷和悬浮电位缺陷局部放电光信号的影响规律。结果表明,随着传播距离的增加:尖刺缺陷检测到光信号最大幅值和放电数均有所衰减,放电谱图变化不大;悬浮缺陷放电光信号幅值明显衰减,放电数不变,放电谱图特征逐渐失去“矩形”特征。此外,当光学信号透过带有通气孔的绝缘子传播时,光学信号明显衰减,且幅值波动较大。

Comparison of the time-domain electromagnetic field from an infinitesimal point charge and dipole source

An electromagnetic field is generated through the accelerating movement of two equal but opposite charges of a single dipole. An electromagnetic field can also be generated by a time-varying infinitesimal point charge. In this study, a comparison between the electromagnetic fields of an infinitesimal point charge and a dipole has been presented. First, the time-domain potential function of a point source in a 3D conductive medium is derived. Then the electric and magnetic fields in a 3D homogeneous lossless space are derived via the relation between the potential and field. The field differences between the infinitesimal point charge and the dipole in the step-off time, far-source, and near-source zones are analyzed, and the accuracy of the solutions from these sources is investigated. It is also shown that the field of the infinitesimal point charge in the near-source zone is different from that of the dipole, whereas the far-source zone fields of these two sources are identical. The comparison of real and simulated data shows that the infinitesimal point charge represents the real source better than the divole source.

Numerical modeling of the 2D time-domain transient electromagnetic secondary field of the line source of the current excitation

To effectively minimize the electromagnetic field response in the total field solution, we propose a numerical modeling method for the two-dimensional （2D） time- domain transient electromagnetic secondary field of the line source based on the DuFort- Frankel finite-difference method. In the proposed method, we included the treatment of the earth-air boundary conductivity, calculated the normalized partial derivative of the induced electromotive force （Emf）, and determined the forward time step. By extending upward the earth-air interface to the air grid nodes and the zero-value boundary conditions, not only we have a method that is more efficient but also simpler than the total field solution. We computed and analyzed the homogeneous half-space model and the fiat layered model with high precision--the maximum relative error is less than 0.01% between our method and the analytical method--and the solution speed is roughly three times faster than the total-field solution. Lastly, we used the model of a thin body embedded in a homogeneous half-space at different delay times to depict the downward and upward spreading characteristics of the induced eddy current, and the physical interaction processes between the electromagnetic field and the underground low-resistivity body.

基于非线性语谱图联合判决的语种识别

针对灰度对数语谱图对基频拉伸幅度过大,短时长语音识别率提升受限的问题,提出一种非线性语谱图联合判决的语种识别方法。首先,对语音进行能量归一化,提取对数功率谱,将频率刻度按照人耳听觉感知进行非线性映射得到非线性语谱图。然后,将非线性语谱图按词关联特性进行等间隔拆分,在ResNet网络后端加入联合判决层;输出语音所属语种类型。实验结果表明,所提方法有效改善灰度对数语谱图的缺点,识别性能均高于语谱图及改进特征。联合判决对切分时长为1.0 s的样本语音取得的识别效果最佳,在广播音频数据集中,识别率达到94.25%;在VoxForge公共语料集中,识别率达到98.94%。

基于二维时频谱图与改进YOLOv5的电能质量扰动识别

随着新型电力系统中新能源渗透率逐渐升高,电网结构复杂性增加,PQD(电能质量扰动)呈现多样化和复杂化的趋势。为实现电能质量扰动的精准识别,提出一种基于二维时频谱图与改进YOLOv5的电能质量扰动图像识别的方法。首先,利用S变换将PQD数据映射成二维时频谱图,通过图像来表征时间、频率和幅值的扰动细节特征;然后,搭建引入ASPP(空洞空间卷积池化金字塔)结构和注意力机制的YO-LOv5训练网络,扩大特征图的感受野以充分提取扰动图像特征,进而以图像识别方法实现PQD分类识别;最后,利用仿真数据进行扰动识别准确率和鲁棒性的验证。结果表明,该方法的识别准确率较高,且图像识别法的引入有助于PQD识别结果的可视化。

基于欠定盲源分离和深度学习的生猪状态音频识别

【目的】为解决群养环境下生猪音频难以分离与识别的问题,提出基于欠定盲源分离与E C A-EfficientNetV2的生猪状态音频识别方法。【方法】以仿真群养环境下4类生猪音频信号作为观测信号,将信号稀疏表示后,通过层次聚类估计出信号混合矩阵,并利用lp范数重构算法求解lp范数最小值以完成生猪音频信号重构。将重构信号转化为声谱图,分为进食声、咆哮声、哼叫声和发情声4类,利用ECA-EfficientNetV2网络模型识别音频,获取生猪状态。【结果】混合矩阵估计的归一化均方误差最低为3.266×10^(−4),分离重构的音频信噪比在3.254~4.267 dB之间。声谱图经ECA-EfficientNetV2识别检测,准确率高达98.35%;与经典卷积神经网络ResNet50和VGG16对比,准确率分别提升2.88和1.81个百分点;与原EfficientNetV2相比,准确率降低0.52个百分点,但模型参数量减少33.56%,浮点运算量(FLOPs)降低1.86 G,推理时间减少9.40 ms。【结论】基于盲源分离及改进EfficientNetV2的方法,轻量且高效地实现了分离与识别群养生猪音频信号。