IDENTIFICATION OF MODAL PARAMETERS FROM MEASURED OUTPUT DATA USING VECTOR BACKWARD AUTOREGRESSIVE MODEL

In this paper, a modal identification system that is based on the vector backward autoregressive (VBAR) model has been developed for the identification of natural frequencies, damping ratios and mode shapes of structures from measured output data. The modal identification using forward autoregressive approach has some problems in discriminating the structure modes from spurious modes. On the contrary, the VBAR approach provides a determinate boundary for the separation of system modes from spurious modes, and an eigenvalue filter for the selection of physical modes is existent in the proposed method. For convenience of application, the backward state equation established from VBAR model is transformed into a forward state equation, which is termed as transformed VFAR model in this paper. In addition, the extraction of equivalent system matrix of state equation of motion for structures from the transformed VFAR model has been developed, and then the normal modes can be calculated from the identified equivalent system matrix. Two examples of modal identification are carried out to demonstrate the availability and effectiveness of the proposed backward approach: (1) Numerical modal identification for a three-degree-of-freedom dynamic system with noise level in 20% of r.m.s of measured output data; (2) experimental modal identification of a cantilever beam. Finally, to show the advantage of the proposed VBAR approach on the selection of physical modes, the modal identification by stochastic subspace method was performed. The results from both methods are compared.     

基于指数加权-核在线序列极限学习机的混沌系统动态重构研究

利用指数加权在线核序列极限学习机(exponential weighted online sequential extreme learning machine with kernel, EW-KOSELM)辨识算法,开展了针对混沌动力学系统的动态重构研究. EW-KOSELM算法将核递归最小二乘(kernel recursive least squares, KRLS)算法直接延伸至在线ELM (extreme learning machine)框架中,通过引入遗忘因子削弱了旧数据的影响,并基于"固定预算(fixed-budget, FB)"内存技术,应对在线核学习算法所固有的规模不断增长的计算困难.将所提辨识算法应用于Duffing-Ueda振子的混沌动力学系统数值仿真实例中,对基于FB-EW-KOSELM的辨识模型与原系统的动态性能进行了定性与定量的分析校验,定性校验准则是基于对比辨识模型与原系统吸引子(轨迹嵌入)、庞加莱映射、分岔图、极限环完成的,定量校验准则包括对比辨识模型与原系统的李雅普诺夫指数与关联维.进一步将其分别应用于来自测量蔡氏电路产生双涡卷吸引子与螺旋吸引子的实测数据实验及某一实际混沌电路所产生的时间序列中,对于具有低信噪比的实测电压或电流数据还需进行了小波降噪预处理.通过分析辨识模型重构吸引子,实验结果表明,FB-EW-KOSELM算法具有良好的动态重构性能,能精确地再生出展示混沌动态行为的过程非线性模型,且具有与原混沌系统非常接近的动态不变性指标.     

Modal identification based on Gaussian continuous time autoregressive moving average model

A new time-domain modal identification method of the linear time-invariant system driven by the non-stationary Gaussian random force is presented in this paper. The proposed technique is based on the multivariate continuous time autoregressive moving average (CARMA) model. This method can identify physical parameters of a system from the response-only data. To do this, we first transform the structural dynamic equation into the CARMA model, and subsequently rewrite it in the state-space form. Second, we present the exact maximum likelihood estimators of parameters of the continuous time autoregressive (CAR) model by virtue of the Girsanov theorem, under the assumption that the uniformly modulated function is approximately equal to a constant matrix over a very short period of time. Then, based on the relation between the CAR model and the CARMA model, we present the exact maximum likelihood estimators of parameters of the CARMA model. Finally, the modal parameters are identified by the eigenvalue analysis method. Numerical results show that the method we introduced here not only has high precision and robustness, but also has very high computing efficiency. Therefore, it is suitable for real-time modal identification.     

Multi-levels inverse identification of physical parameters of porous materials

The identification of the physical parameters of porous materials presents an important field of research, in which many identification methods are developed. One of those methods is presented in this paper. In fact, a multi-levels inverse identification method is developed in order to estimate these physical parameters. The proposed method is based on the minimization of the difference between a reference acoustic absorption coefficient of a porous material and the computed values. The minimization is done according three levels, in each level an acoustic model of porous material is used to evaluate one or two parameters. Finally, the five physical parameters of the porous materials are deduced. The proposed method is applied to Polyurethane foam material. The obtained results are satisfying with small values of errors and with estimated acoustic absorption coefficient reaching the reference one.     

Flexible polyurethane foam modelling and identification of viscoelastic parameters for automotive seating applications

A hereditary model and a fractional derivative model for the dynamic properties of flexible polyurethane foams used in automotive seat cushions are presented. Non-linear elastic and linear viscoelastic properties are incorporated into these two models. A polynomial function of compression is used to represent the non-linear elastic behavior. The viscoelastic property is modelled by a hereditary integral with a relaxation kernel consisting of two exponential terms in the hereditary model and by a fractional derivative term in the fractional derivative model. The foam is used as the only viscoelastic component in a foam-mass system undergoing uniaxial compression. One-term harmonic balance solutions are developed to approximate the steady state response of the foam-mass system to the harmonic base excitation. System identification procedures based on the direct non-linear optimization and a sub-optimal method are formulated to estimate the material parameters. The effects of the choice of the cost function, frequency resolution of data and imperfections in experiments are discussed. The system identification procedures are also applied to experimental data from a foam-mass system. The performances of the two models for data at different compression and input excitation levels are compared, and modifications to the structure of the fractional derivative model are briefly explored. The role of the viscous damping term in both types of model is discussed.     

Characteristic model of travelling wave ultrasonic motor

In general, the design and analysis of ultrasonic motor and motor’s control strategy are based on mathematical model. The academic model is widely used in the analysis of traveling wave ultrasonic motor (TWUSM). But the dispersive characteristic of piezoelectric ceramics and other complicated process, such as the friction, make the model’s precision not so accurate. On the other hand, identification modeling method, which is built based on the tested data, has obtained increasing application in the study of ultrasonic motor’s control technology. Based on the identification model, many control strategies can be designed easily. But the identification model is an approximate model, so if a more accurate model of ultrasonic motor can be obtained, the analysis and design of motor control system will be more effective. Characteristic model is a kind of identification model which can accurately describe the characteristics of TWUSM. Based on the tested data, this paper proposes the modeling method of ultrasonic motor’s characteristic model. The paper also makes a comparison of the effectiveness of different identification algorithms. Aiming at the speed control of ultrasonic motor, the influence of the parameter’s initial values on the precision of model is discussed. The calculating results indicate the availability of this characteristic model.     

基于长短时记忆神经网络的能谱核素识别方法

针对新兴的能谱核素识别方法在混合放射性核素的噪声环境中存在识别速度慢、准确率较低等问题,提出了基于长短时记忆神经网络(LSTM)的能谱核素识别方法。实验使用溴化镧(LaBr3)晶体探测器,分别对环境中60Co、137Cs放射性源分组测量得到能谱数据集,首先使用数据平滑方法和归一化方法进行数据预处理,然后将能谱数据按时间序列分组以获得可用的输入序列数组,最后训练LSTM模型得到预测结果。通过基于BP神经网络和卷积神经网络(CNN)的两个能谱识别模型进行对比,得到在测试集中平均识别率分别为83.45%和86.21%,而LSTM能谱识别模型平均识别率为93.04%,实验结果表明,该能谱模型在核素识别效果中表现较好,可用于快速的能谱核素识别设备上。     

脉冲电流探头的时域标定研究

 介绍了一种Rogowski线圈脉冲电流探头时域标定和标定数据处理的方法,将脉冲法时域标定的结果和频域标定结果进行了比较。针对电流探头的低频失真,采用系统辨识方法建立了测量系统的动态模型,并对探头输出信号的失真进行校正。实测数据验证了所建模型的有效性,设计出的数字补偿滤波器可将测量系统校正为一理想的比例环节。     

基于裂变γ标识技术的瞬发裂变中子谱测量新方法

瞬发裂变中子谱（prompt fission neutron spectrum,PFNS）是用于核实验诊断过程中十分重要的参数数据,传统的测量主锕系核素（U,Pu）PFNS的技术手段是采用裂变室,利用裂变碎片标识裂变中子,通过中子飞行时间技术获得裂变中子谱.目前出现了一种新的用于PFNS测量的技术,其原理是基于如下的物理事实：在一次裂变过程中,释放中子的同时伴随着释放7–8个γ射线光子,而非弹性散射效应产生的γ射线光子只有1–2个.据此,可以通过裂变γ射线的多重性将裂变中子和其他杂散中子甄选出来,达到测量PFNS的目的.本文建立了基于裂变γ标识技术的PFNS测量实验系统.利用该系统对²⁵²Cf中子源的PFNS进行了实验测量,测量结果与传统的裂变碎片标识法及ENDF/B-VⅡ数据库的标准谱进行了比较,对新方法的裂变标识率以及实验不确定度也一并进行了分析.     

Discrete state space method and modal extension method based impact sound synthesis model

The efficient and accurate synthesis of physical parameter-controllable impact sounds is essential for sound source identification. In this study, an impact sound synthesis model of a cylinder is proposed based on discrete state space(DSS) method and modal extension method(MEM). This model is comprised of the whole three processes of the physical interaction, i.e., the Hertz contact process, the transient structural response process, and the sound radiation process. Firstly,the modal expanded DSS equations of the contact system are constructed and the transient structural response of the cylinder is obtained. Then the impact sound of the cylinder is acquired using improved discrete Raleigh integral. Finally, the proposed model is verified by comparing with existing models. The results show that the proposed impact sound synthesis model is more accurate and efficient than the existing methods and easy to be extended to the impact sound synthesis of other structures.     

基于COMPASS+PD/HRLR组合的侦察导弹对海洋移动目标定位方法研究

针对现行侦察手段对敌方海洋移动目标进行地理坐标的测量方法不多,测量精度不高的问题,提出了一种利用反舰巡航导弹搭载北斗卫星定位接收机(COMPASS)、合成孔径雷达(SAR)、脉冲多普勒雷达(PD)、高重频激光测距仪(HRLR)、激光测高仪(LHD)与数据传输系统(DCS),实现对敌海洋移动目标准确识别和精确定位的方法,介绍了侦察导弹的组成,说明了定位原理,给出了物理多站连续交会定位算法模型,进行了模拟测量数据解算和误差分析。通过仿真证明,本方法简单、实用,能够满足实际工程应用的需求,可为海上远程精确打击体系提供高精度的目标位置信息。     

PREDICTION OF DYNAMIC CHARACTERISTICS USING UPDATED FINITE ELEMENT MODELS

Model updating techniques are used to update a finite element model of a structure so that an updated model predicts the dynamics of a structure more accurately. The application of such an updated model in dynamic design demands that it also predict the effects of structural modifications with a reasonable accuracy. This paper deals with updating of a finite element model of a structure and its subsequent use for predicting the effects of structural modifications. Updated models have been obtained by a direct model updating method and by an iterative method of model updating based on the frequency response function (FRF) data. The suitability of updated models for predicting the effect of structural modifications is evaluated by some computer and laboratory experiments. First a study is performed using a simulated fixed-fixed beam. Cases of complete, incomplete and noisy data are considered. Updated models are obtained by the direct and the FRF-based method in each of these cases. These models are then used for predicting the changes in the dynamic characteristics brought about due to a mass and a beam modification. The simulated study is followed by a study involving actual measured data for the case of an F-shape test structure. The updated finite element models for this structure are obtained again by the direct and the FRF-based method. Structural modifications in terms of mass and beam modifications are then introduced to evaluate the updated model for its usefulness in dynamic design. It is found that the predictions based on the iterative method based updated model are reasonably accurate and, therefore, this updated model can be used with reasonable accuracy to perform dynamic design. The predictions on the basis of the direct method based updated model are found to be reasonably accurate in the lower portion of the updating frequency range but the predictions are in a significant error in the remaining portion of the updating frequency range. It is concluded that the updated models that are closer to the structure physically are likely to perform better in predicting the effects of structural modification.     

基于多类型光谱特征参数匹配的矿物信息遥感识别方法

传统的与光谱反射率匹配的矿物填图方法受图像质量、大气及环境背景等因素影响较为明显,矿物识别效果较差。针对上述问题提出一种基于多类型光谱特征参数匹配的矿物信息识别方法,综合利用多种光谱特征参数在大气及环境背景变化中的相对稳定性特点,实现矿物信息的高精度识别。选取美国内华达州Cuprite矿区AVIRIS机载可见-红外光谱成像仪高光谱数据,以USGS美国地质调查局波谱库中的矿物波数为参照光谱,分别计算AVIRIS影像和USGS波谱库中典型矿物的光谱特征参数,在对多种光谱特征参数综合分析的基础上,构建Cuprite矿区矿物识别模型,进行矿物填图实验。最后使用Clark等在该地区的矿物填图结果数据,对实验结果进行了验证,结果表明：该方法矿物识别结果与Clark等的填图结果吻合度高,整体矿物识别精度达到78.96%。     

空间目标可见光散射特性建模方法研究

针对空间目标的可见光散射特性提出一种建模方法.在分析空间目标所处的背景辐射环境基础上建立了空间目标背景辐射物理模型.对目标表面进行面元划分后,基于辐射理论引入双向反射分布函数模型来描述目标表面面元的光散射特性,将目标各个表面所有面元散射分量叠加建立了目标可见光散射特性的数学模型.建立目标本体坐标系,通过坐标变换确定目标、背景辐射源与探测器的相对位置关系,利用矢量坐标法确定目标对观测系统的“可视表面”.根据给定的目标几何结构尺寸和物性参量仿真获得了目标在轨光学特性.计算结果验证了建模的有效性.     

基于拉曼光谱和机器学习的百合分类识别

百合鳞茎,百合科百合属多年生草本球根植物生长的肥厚鳞片构成的地下变态茎,是一种典型的药食同源作物,含有丰富营养成分的同时还具有抗肿瘤、抗抑郁、降血糖、提高免疫力等保健作用。不同产地百合鳞茎的市场价格差异较大,依赖于人工经验感官的传统评测方法主观性强、确定性差,难以广泛推广在现代生产环节。以化学检验法为主的先进检测方法耗时长、费用高,而且难以满足产地鉴别这一要求。针对百合鳞茎现场快速地产地判断和品质评价的需求,提出了一种使用拉曼光谱和机器学习的百合鳞茎检测方法。拉曼光谱是一种基于非弹性散射的振动光谱,能够做到快速准确的无损检测,将拉曼光谱与机器学习算法相结合,建立了我国分布最为广泛的三种百合鳞茎(兰州百合、宜兴百合和龙牙百合)的产地分类模型,着眼于基质光谱上479, 870, 942和1 606cm^-1等特征峰,提出了一种基于拉曼光谱的成分含量判断产地和评价百合鳞茎品质的无损检测方法。首先采集百合鳞茎样本的光谱,经过光谱数据预处理后,使用人工先验法提取百合鳞茎代表物质并确定特征峰,再使用主成分分析和t-分布随机邻域嵌入方法降维提取光谱数据特征。并将获得的数据特征分别...     

Substructure methods for structural condition assessment

It is commonly known that an accurate analysis of a large structure requires an accurate analytical model. This is also true for the inverse analysis of a structural system where measured structural responses are used as input to assess the structural conditions. However, an accurate model of the structure is always not available in practice. Two substructural identification methods are presented in this paper with the structure divided into substructures and with one substructure assessed at one time. In the first method, an accurate finite element model of the whole structure is assumed known. A state space method is applied to identify the external forces acting on the structure, and a damage identification method is then applied to identify the local damages using time domain information. Iterative model updating method based on the measured acceleration in the selected substructure is employed for the assessment. The second identification method requires only the finite element model of the substructure. The interface forces and the external forces acting on the target substructure are all taken as excitations and they are identified in state space. The substructure is then assessed similar to the first method. Since the target substructure for updating consists of a much reduced number of components and the identification problem is more efficient. The validation of the proposed methods is demonstrated by a truss structure with polluted measured accelerations with promising results.     

优化模式物理参数的扩展四维变分同化方法

数值模拟的一个重要误差来源是模式物理参数,为提高模拟准确率,如何改进模式物理参数是亟需解决的问题.本文对经典四维变分同化技术进行了改进,提出了一种新的利用观测资料来同时优化模式初始场和物理参数的扩展四维变分同化方法,并以Ekman边界层模式和Lorenz模式为例进行了数值试验.结果表明,利用本文提出的新方法,通过对观测资料的变分同化,可以在实现对模式初始场进行优化的同时,纠正了模式物理参数中的误差,从而有效提高了模式的模拟准确率.该方法对于改进数值模式物理参数有着重要的促进意义.     

具有迟滞非线性的金属橡胶隔振器参数识别研究

金属橡胶隔振器具有非线性的动力学特性,对这种迟滞阻尼隔振器进行建模研究,建立了参数有物理意义的动力学模型.根据单自由度性能试验,进行了有关参数的试验识别方法研究,应用能量法及最小二乘法将非线性方程组转化为关于参数的线性方程组,从而对金属橡胶隔振器参数进行识别.识别结果与实验结果有较好的一致性,识别结果也说明了结构参数对隔振器性能的影响. 关键词： 金属橡胶隔振器 动力学模型 迟滞非线性 参数识别     

基于反贝叶斯学习的WDMS光谱自动识别研究

天体光谱是天体物理学重要的研究对象,通过光谱可以获取天体的许多物理、化学参数如有效温度、金属丰度、表面重力加速度和视向速度等。白矮主序双星是一类致密的双星系统,对研究致密双星的演化特别是公共包层的演化有着重要的意义。国内外的大型巡天望远镜如美国斯隆望远镜以及中国的郭守敬望远镜,每天都产生大量光谱数据。如此海量的光谱数据无法完全用人工进行分析。因此,使用机器学习方法从海量的天体光谱中自动搜索白矮主序双星光谱,有着非常现实的意义。目前的光谱自动识别方法主要通过对已有的标签样本进行分析,通过训练得到分类器,再对未知目标进行识别。这类方法对样本的数量有明确的要求。白矮主序双星的实测光谱数量有限。若要通过有限的样本集准确学习白矮主序双星的光谱特征,不仅需要扩大样本数量,还需要提高特征提取和分类算法的精度。在前期工作中,通过机器学习等方法在海量巡天数据中识别了一批白矮主序双星的光谱,为该实验提供了数据源。使用对抗神经网络生成新的白矮主序双星光谱,扩大训练数据量至原数据集约两倍的数量,增强了分类模型的泛化能力。通过反贝叶斯学习修正损失函数,将损失函数的大小与样本的方差相关联,抑制了异常数据对模型造成的影响,提升了模型的鲁棒性,解决了由于训练样本集偏差带来的梯度消失以及训练陷入局部最优解等问题。该实验基于Tensorflow深度学习库。使用Tensorflow搭建的生成对抗网络具有较好的鲁棒性,并且封装了内部实现细节,使得算法得以更好地实现。除此之外,由Tensorflow搭建的卷积神经网络在该实验中用于分类准确度测试。实验结果表明,二维卷积神经网络能够利用卷积核有效地提取白矮主序双星的卷积特征并进行分类。基于反贝叶斯学习策略的卷积神经网络分类器在白矮主序双星原始数据及对抗神经网络生成光谱的识别任务中达到了约98.3%的准确率。该方法也可用于在巡天望远镜的海量光谱中搜索其他特殊和稀少天体如激变变星、超新星等。     

基于粒子滤波状态估计的滚动轴承故障识别方法

提出了一种基于粒子滤波状态估计的滚动轴承故障识别方法,该方法主要包括故障模型建立和故障识别两个步骤。在故障模型建立部分,首先依据滚动轴承不同故障状态下的振动信号,建立对应的自回归模型,作为故障模型;在故障识别部分,将正常状态下对应的模型,转化为状态空间模型,设计粒子滤波器,然后对不同的故障状态进行估计,提取其残差的相关特征,并结合模型参数特征应用BP神经网络识别算法进行故障识别。最后以美国凯斯西储大学的滚动轴承振动数据为例,验证了该方法的有效性。