Aligning national “logics” in a European military helicopter program

This article analyzes the public side of the NH90 network consisting of four participating countries (Germany, France, Italy, and the Netherlands) and their industrial partners. Comparable to observations in earlier international projects in the defense sector, the development and production of the NH90 defense helicopter did not match original plans and costs estimates. On the basis of four mechanisms that were intended to facilitate the cooperation between the partnering countries (the General Memorandum of Understanding; coalition formation; the role of the central agency; and the process of vertical escalation) the question is posed whether or not there was a true alignment of national interests and “logics.” The lack of standardization, as well as program delays and issues concerning the division of the work-share, lead to the emerging alignment being characterized as “quasi-alignment” at best.     

钢质油罐罐底腐蚀预测模型研究与评价

钢质油罐在储存油料时不可避免地会受到严重腐蚀.为能够掌握罐底的腐蚀程度,提出了一种由灰色系统理论和BP人工神经网络相结合的方法建立预测模型,对钢质油罐罐底腐蚀情况进行了预测,并通过实例计算给出了评价.计算结果表明,预测值与实测值相差很小,尤其对油料腐蚀这种涉及较多因素的复杂过程,该方法具有明显的优越性;该模型克服了传统预测模型需建立函数的难题,且预测精度较高,具有较高理论与实际应用价值.     

基于小波神经网络的货运量预测

选取我国1979—2011年国民生产总值、能源生产总值、进出口贸易额、社会消费品零售总额和固定资产投资总额5项指标作为货运总量的主要影响因素,以货运总量为输出,建立了基于Morlet小波函数的小波神经网络预测模型。该模型能够揭示货运量与相关变量之间的非线性映射关系,经实例分析得到了较满意的结果,并通过与实际货运量和BP神经网络预测结果的对比,证明了小波神经网络在货运量预测方面应用的可行性。     

基于复杂网络的指控组织建模与分析

面对指控组织(C2组织)日益呈现出的复杂性和网络化结构特点,提出基于复杂网络的研究思路,并从现实指控组织网络出发,分析复杂网络理论适用于指控组织建模的现实与理论依据。结合指控组织建设的普遍实际,参考复杂网络适应度模型及组织网络模型后,基于新节点和内部边两方面交互增长的思想,建立了指控组织生成模型。通过仿真实验,验证了无标度性、小世界性等复杂网络特性,符合现实指控组织的特点,为进一步进行指控组织效能评估提供了基础。     

压电执行器 NLDDS及神经网络非线性建模与控制

本文采用非线性动态数据系统(NLDDS)建模方法和非线性递推最小二乘算法,对具有迟滞非线性特性的压电陶瓷微进给执行器的动态特征进行了建模、预报与控制,并且用人工神经网络对该类系统的建模与控制进行仿真。结果证明上述方法在一定程度上是可行的和有效的。     

利用多时间尺度卷积的视频行为识别网络

基于2D的行为识别网络通常融合多张视频帧的分类结果识别不同的行为,但其在卷积过程中缺少对时空特征提取。针对该问题,基于时间位移模块(temporal shift module,TSM)的思想设计了一组多时间尺度卷积,包含不同设计的卷积核以提取融合不同时间尺度的时空信息。通过控制多时间尺度卷积嵌入ResNet50网络的位置及其模块的参数设置,寻找最优的基于多时间尺度卷积的行为识别网络。使用PyTorch深度学习框架训练模型,在大型开源数据集Something-Somethingv2上进行了实验研究。结果表明,基于多时间尺度卷积的行为识别网络对行为识别准确率达到了59.47%,优于TSM等网络。     

基于PSO-RBF神经网络的雷达目标识别

为提高雷达系统目标识别能力,对粒子群算法及RBF神经网络进行了分析。针对离子群算法(PSO)易陷入局部极小的缺陷,提出了基于自适应时变权重和局部搜索算子的改进PSO算法,并将该算法应用到RBF神经网络核函数参数的优化学习中,进行了雷达目标识别仿真实验。仿真结果表明,相对于标准PSO-RBF神经网络,改进算法不仅收敛速度快,且误差精度高,特别在干扰较强时,目标的识别率有较大提高。     

基于CNN的三相逆变器开路故障诊断及其样本条件分析

为了合理选择样本条件以实现高效的智能化诊断,以及克服智能化方法中传统反向传播(back propagation, BP)网络权值较多、局部信息提取能力不足的问题,对基于卷积神经网络(convolutional neural network, CNN)的开路故障诊断方法进行研究,并以典型的三相两电平逆变器为具体对象,着重分析样本时长、样本数量变化时,CNN方法相较于BP网络方法在网络权值数量、训练稳定性、诊断准确率上的量化优势。结果表明,基于CNN的方法可在权值数量远少于BP网络方法的情况下构建深度更深的诊断模型,并在更短样本时长、更少训练样本数量下实现高效、准确的开路故障诊断。     

Data-driven prediction of plate velocities and plate deformation of explosive reactive armor

Explosive reactive armor (ERA) is currently being actively developed as a protective system for mobile devices against ballistic threats such as kinetic energy penetrators and shaped-charge jets. Considering mobility, the aim is to design a protection system with a minimal amount of required mass. The efficiency of an ERA is sensitive to the impact position and the timing of the detonation. Therefore, different designs have to be tested for several impact scenarios to identify the best design. Since analytical models are not predicting the behavior of the ERA accurately enough and experiments, as well as numerical simulations, are too time-consuming, a data-driven model to estimate the displacements and deformation of plates of an ERA system is proposed here. The ground truth for the artificial neural network (ANN) is numerical simulation results that are validated with experiments. The ANN approximates the plate positions for different materials, plate sizes, and detonation point positions with sufficient accuracy in real-time. In a future investigation, the results from the model can be used to estimate the interaction of the ERA with a given threat. Then, a measure for the effectiveness of an ERA can be calculated. Finally, an optimal ERA can be designed and analyzed for any possible impact scenario in negligible time.     

基于网格划分的无线传感器网络多重覆盖算法

为了延长无线传感器网络的工作周期,在满足网络覆盖性能的前提下,可利用调度算法让一部分节点进入休眠以节省能量。提出了一种基于网格划分的无线传感器网络多重覆盖算法,新算法包括冗余节点判断和节点调度两部分。将节点覆盖区域划分为多个网格,通过判断各个网格是否满足覆盖要求,进而判断节点是否冗余。新算法给出了边界冗余节点判据,在调度过程中能够克服边界效应的影响,同时通过冗余节点能量比较,避免了休眠冲突和覆盖盲区的产生。仿真结果表明,与传统的CPNSS算法相比,新算法对冗余节点的判断更为准确,在网络工作集和平均覆盖度两项性能评价指标上均优于传统调度算法,且对网络节点数量增加造成的影响不敏感,能够有效地减少网络冗余,起到了提升网络性能的效果。