基于代理模型进化的履带车辆动力学参数优化

为解决履带车辆动力学优化过程因代理模型构建和应用的不足而导致的参数寻优精度和效率不高的问题,提出基于代理模型进化的参数优化方法,将优化迭代和代理模型动态构建相融合,来降低仿真模型调用次数和提高优化效率。根据车辆几何拓扑结构,构建考虑履带包络效应的车辆多体动力学模型;提出优化设计变量空间的3层子空间划分方法,以及具有空间聚焦、空间约简和跳出局部寻优的多级模糊聚类空间约简方法,实现设计参数在3层子空间的高效缩减。以履带车辆多体动力学模型参数优化过程为例进行应用验证。研究结果表明：在3种路况下的履带车辆多体动力学优化过程降低仿真模型调用次数最高可达85%;表征履带车辆行进特性的综合性能指标分别提升约32.4%、24.5%、20.4%。证明了新方法可有效提高动力学模型的优化效率和精度。     

轻武器射击系统建模及计算结果分析

通过简化臣姿持枪射击系统，利用拉格朗日方程，建立了该射击系统的物理模型和数学模型，并通过数值模拟，论述了缓冲簧对系统动态特性的影响，最后通过实验，对无缓冲簧射击系统的动态特性参数进行验证。     

Response of SDOF bilinear elasto-dynamical models of cracked concrete beams for El Centro earthquake

Investigated herein are the dynamic responses of three different single degree of freedom (SDOF) bilinear dynamical models of a cracked concrete beam subjected to the El Centro earthquake. The first model is an equivalent unilinear model while the other two models have different stiffness values but differ from each other in the dependence of their damping coefficients on the sense of amplitude. All three models have the same natural period. The three models have been found to predict different response spectra for the same bilinear dynamical system. The effect of constant loading on the dynamical response of the concrete structures subjected to sinusoidal and random loading has been found to be significant for more flexible structures. Implications of these theoretical predictions for seismic design of concrete structures are discussed.     

液压/燃油管道系统的管径及其支撑的动力优化设计

建立了有、无弹性刚度支撑的液压/燃油管道系统,采用了动力优化的有限元分析方法,分析了在一个频率约束下以质量最轻为目标的管道系统,得出了管径、支撑刚度和固有频率的定量关系,提出了防止管道系统强迫共振的方法,最终为管道系统的安全性指导和设计提供了理论依据。     

分形几何和动力系统中的一些问题

本文是Zhou与Feng,Nonlinearity,2004的续篇,对自相似集的结构和动力系统提出一些较为深入的问题。     

航空发动机润滑油光谱分析界限值动态调整问题研究

将航空发动机润滑油光谱分析的界限值作为动态系统进行研究,对传统的制定界限值的方法进行了对比研究,运用神经网络理论,结合故障诊断思想,建立了动态界限值模型;针对动态界限值模型泛化能力不高的问题,建立遗传神经网络模型,实现了动态界限值模型的优化;利用积累的航空发动机运行的故障数据和实际工作中获得的数据进行分析,验证了动态界限值模型的有效性.     

定常约束力学系统的相空间上的联络

提出动力学流关于切丛上联络的一种协变性，求出定常约束力学系统的联络，其测地线方程为系统的运动方程，该联络由系统的动力学函数和约束确定且当动力学流为二阶齐次时约化为通常的动力学联络。     

战场宽带数据链分布式虚拟骨干网的构建

宽带数据链作为现代战场的神经网络和信息传输通道,必须采用分布式虚拟骨干网构建算法才能适应战场环境的大容量、多样性数据传输和时变性网络结构。针对这一问题,提出了一种分布式虚拟骨干网构建算法-DCDS算法。该算法中,每个节点只需获取其两跳范围内的邻居节点信息,无需获知全网拓扑信息。理论分析和仿真表明,相比Wu等2种经典算法,DCDS算法具有更小的消息开销和虚拟骨干网构建规模,更适合于大数据量、高动态的战场宽带数据链网络。     

Validation of dynamic cavitation model for unsteady cavitating flow on NACA66

Unsteady cavitating flow is extremely complicated and brings more serious damages and unignorable problems compared with steady cavitating flow.CFD has become a practical way to model cavitation;however,the popularly used full cavitation model cannot reflect the pressure-change that the bubble experiences during its life path in the highly unsteady flow like cloud cavitating.Thus a dynamic cavitation model(DCM)is proposed and it has been considered to have not only the first-order pressure effects but also zero-order effect and can provide greater insight into the physical process of bubble producing,developing and collapsing compared to the traditional cavitation model.DCM has already been validated for steady cavitating flow,and the results were reported.Furthermore,DCM is designed and supposed to be more accurate and efficient in modeling unsteady cavitating flow,which is also the purpose of this paper.The basic characteristic of the unsteady cavitating flow,such as the vapor volume fraction distribution and the evolution of pressure amplitude and frequency at different locations of the hydrofoil,are carefully studied to validate DCM.It is found that not only these characteristics mentioned above accord well with the experimental results,but also some detailed transient flow information is depicted,including the re-entrant jet flow that caused the shedding of the cavity,and the phenomenon of two-peak pressure fluctuation in the vicinity of the cavity closure in a cycle.The numerical results validate the capability of DCM for the application of modeling the complicated unsteady cavitating flow.     

基于动态优化PCA人脸识别算法的实现

提出一种人脸识别动态优化PCA算法。采用ORL人脸数据库,对数据库中图像进行放缩归一化操作并按升序或降序排列。然后,将数据库分为最大值类、中值类、最小值类三个部分,求其各类平均值,特征值个数选取9和10,实现人脸识别和重建。随机采样人脸图像,采用DCT算法将其转换到频域进行分析,通过比较整个数据库平均脸与三类中各自平均脸,算法运行后,特征值个数可以实现自动优化。实验结果显示,该算法在一定程度上可用来对传统PCA算法部分关键参数进行优化。