基于神经网络的机械磨损故障光谱定位诊断法

在分析常用光谱定位诊断方法的基础上提出了基于神经网络的光谱定位诊断法；将机械摩擦副材质的元素含量作为神经网络输入，将材质所对应的部件作为神经网络输出，建立了相应的神经网络训练样本；通过整理训练样本和训练神经网络，利用神经网络超强的非线性映射能力和容错性实现了磨损故障部位诊断；通过算例分析验证了所提出的诊断方法的可行性和准确性．结果表明，所建立的方法简洁有效，并具有很高的诊断精度．     

柴油机磨损状态监测及故障诊断专家系统知识库建立的研究

实践表明，为了确保机器的安全运行，避免突发性故障，对其实施磨损状态监测和故障诊断是十分必要的。尽管铁谱技术和光谱技术等在这方面的应用都比较成功，然而它们也各有其局限性。为了把机器的磨损状态监测和故障诊断技术推向更高的科学水平，在摩擦学系统的系统工程思想指导下，针对柴油机的“拉缸”和“拉瓦”故障，并且根据故障发生之前润滑油结构变化的各种信息，利用专家系统开发工具ＥＸＳＹＳ，建立了基于铁谱技术和光谱技术的柴油机磨损状态监测及故障诊断专家系统ＦＤＥＸＳＹＳ（Ｖ１．０）知识库；对该专家系统的测试和实际运行结果表明，所建立的知识库可以比较全面而可靠地描述柴油机的磨损状态及故障前兆，因而能够提醒用户适时采取有效措施以防止故障的发生。不仅如此，所述专家系统知识的获取方法、知识的分类表示及匹配方法等研究，对建立其它摩擦学系统的故障诊断专家系统都具有参考意义。     

国内外 INS 和 SG 故障诊断专家系统的发展概述

本文着重介绍了几种典型的专家系统，如国内ＰＬ—１型平台罗经自动测试系统（ＰＬＴＳ），国内外ＤＭＩＮＳ的复合诊断系统（ＢＤＳ）以及Ｆ—１５飞机惯导智能诊断专家（ＩＮＳＩＤＥ）等；并剖析其技术特点，说明采用专家系统诊断ＩＮＳ和ＳＧ故障的必要性。     

基于减聚类和T-S模糊神经网络的转台故障诊断系统

在分析转台故障类型和机理的基础上,针对故障与征兆之间的复杂非线性映射关系,提出了一种基于减聚类的T-S型模糊神经网络故障诊断方案.首先建立了转台故障底事件与征兆信号的明确对应关系,并以清晰数值形式表述专家的诊断经验;然后在减聚类算法中引入权值的概念,获得简约规则表;接着利用抗噪声训练方法训练网络,使其能够克服一定幅值内的噪声干扰;最后利用含噪声数据和测试数据分别试验.试验结果表明:该方法能有效减少诊断规则的数目,准确地实现故障识别,对噪声的容错能力强,有很强的工程实用性.     

智能Kalman滤波在水下地形组合导航系统中的应用

为了避免现实环境的动态变化对水下航行器系统模型造成的随机干扰影响,保证水下航行器长时间导航精度的稳定性,提出了利用RBF神经网络辅助联邦Kalman滤波方法对SINS/TAN/DVL/MCP组合导航系统进行信息融合。给出了各子导航系统的误差模型,通过足够精度的样本对前向神经网络进行离线训练,建立神经网络控制模型。仿真结果表明,该方法可使水下航行器的系统状态在较短的时间内以较高的精度达到稳定。通过与联邦Kalman滤波结果对比表明,采用智能控制方法辅助的信息融合方式的导航定位精度提高了一倍,能有效提高常规联邦Kalman滤波器的自适应能力,达到减小误差,提高精度的目的。     

转台的粗糙神经网络故障诊断系统设计

针对转台故障的多样性与复杂性,设计了独立于专家的粗糙神经网络故障诊断系统。首先建立转台故障诊断决策表,然后用粗糙集方法约简冗余属性,最后设计了神经网络分类器和辨识器。实验结果显示,诊断系统能较好地区分和辨识具有相同故障现象的不同故障,诊断正确率达到96.7%。将粗糙集理论与神经网络相结合,简化了信息表达空间,减小了神经网络结构的复杂性,并具有强大的容错和抗干扰能力,工程实用性强。     

基于遗传RBF网络的惯性测量组合模拟电路故障诊断

为解决惯性测量组合模拟电路的诊断不易定位到元件级故障的问题，提出了一种基于遗传RBF网络的智能诊断方法。该方法首先利用RBF神经网络快速准确识别故障的能力，以RBF的训练均方误差为遗传算法的适应度函数，依靠遗传算法强大的全局寻优能力实现故障特征选择。在特征选择的过程中，同时记录使训练均方误差达到最小的最优RBF网络，然后直接利用特征选择过程中训练好的最优RBF网络诊断故障，而无需利用特征选择后的训练数据对RBF网络进行再训练，简化了诊断步骤，同时增强了网络的抗干扰能力。仿真结果表明，该方法能有效去除冗余特征，准确诊断惯性测量组合模拟电路的故障，并有较好的抗噪能力，证明了该方法的有效性和可行性。     

基于遗传算法优化神经网络的多源信息融合室内定位方法

为了解决复杂室内环境中单一定位技术误差较大的问题,提出了一种基于遗传算法优化BP神经网络的多源信息融合室内定位方法。首先利用Wi Fi定位结果约束地磁匹配范围进行组合定位,降低误匹配率;再采用遗传算法寻找网络全局最优解对BP神经网络的初始权值和阈值进行优化,提升网络精度并加快收敛;使用优化后的网络对组合定位结果和推算定位结果向真实位置坐标方向训练融合,得到最优定位结果。数据显示,经遗传算法优化后BP神经网络预测均方误差降低了约75%,融合定位精度较单一定位方式定位精度平均提升约47%。结果表明,所提的方法可有效提升定位精度,具有更优的定位性能。     

基于人工神经网络的声子晶体逆向设计

声子晶体是一种人工周期性复合材料, 其带隙特性使其在减振、隔声、滤波和声学功能器件等领域具有潜在的应用价值. 如何准确操纵声波和机械波是声子晶体设计的主要挑战. 现有设计方法是基于对结构几何参数与材料参数的分析调整使其匹配特定的应用特性, 设计效率不高且无法达到最佳性能. 为此, 本文以一维层状声子晶体为例, 提出了一种基于Softmax逻辑回归和多任务学习的人工神经网络声子晶体逆向设计方法, 其中, Softmax逻辑回归实现分层结构各区域材料种类的选择, 通过多任务学习确定各区域材料的分布, 从而, 将声子晶体逆向设计问题转化为对单位胞元拓扑结构多组分材料的分类问题. 首先, 随机生成大量声子晶体拓扑结构样本; 然后, 采用有限元法进行并行计算得到所有样本的带隙分布; 接着, 通过神经网络建立带隙分布和拓扑结构之间的映射关系; 最后, 利用训练好的神经网络设计具有目标带隙特性的声子晶体, 即以目标带隙作为神经网络的输入, 网络将直接输出对应的声子晶体单元胞元拓扑结构. 算例表明本方法可根据应用需求快速高效地得到具有目标带隙的一维声子晶体. 该方法为声子晶体的逆向设计提供了一种新颖思路.      

基于模糊神经网络的故障诊断及其在无刷直流电机系统中的应用

本文针对故障诊断知识具有模糊性的特点,提出了一种基于模糊神经网络的故障诊断方法,将模糊理论与神经网络融合在一起,实现了对故障的模糊诊断.文中给出了模糊神经网络的结构和学习方法,并提出了一种阈值向量故障判别方法,使故障判别更具灵活性.并以无刷直流电机系统为对象,将该方法应用于系统的故障诊断,仿真结果表明,该方法是行之有效的.     

Preface: Current research progress on mechanics of high speed rail

The construction and operation of high-speed rail（HSR）grid within the past two decades in China,in terms of the scale,may be possibly comparable to any national-wide construction in the history of China,even the Great Wall.By counting railways with commercial train service at the speed of200 km/h,China has the world’s longest HSR network with over 19 369.8 km of track in service today and this number     

ACTA MECHANICA SINICA 2014 ANNUAL SUBJECT INDEX

正~~     

人造地震动合成中的位移误差分析

三角函数级数法是合成人工地震动常用算法之一,但是通过对加速度积分求取位移时,却存在与零线漂移相类似的位移漂移现象。     

Hermitian space-time finite elements for estuarine mass transport

Space-time finite element solutions of the convection–dispersion equation using higher-order nodal continuity and Hermitian polynomial shape functions are described. Five separate elements ranging from a complete linear element with C^0,0 nodal continuity to a complete first-order Hermitian element with C^1,1 nodal continuity are subjected to detailed analysis. Wave deformation analyses identify the source of leading or trailing edge oscillations, trailing edge oscillations being the major source of difficulty. These observations are confirmed by numerical experiments which further demonstrate the potential of higher-order nodal continuity. The performance of the complete first-order Hermitian element is quite satisfactory and measurably superior to the linear element.     

Rainwater rivulets on a cable subject to windFilets d'eau de pluie sur un câble soumis au vent

Rainwater rivulets appear on inclined cables of cable-stayed bridges when wind and rain occur simultaneously. In a restricted range of parameters this is known to cause vibrations of high amplitudes on the cable. The mechanism underlying this effect is still under debate but the role of rainwater rivulets is certain. We use a standard lubrication model to analyse the dynamics of a water film on a cylinder under the effect of gravity and wind load. A simple criterion is then proposed for the appearance and position of rivulets, where the Froude number is the control parameter. Experiments with several geometries of cylinder covered with water in a wind tunnel show the evolution of the rivulets with the Froude number. Comparison of the prediction by the model with these experimental data shows that the main mechanism of rivulet formation and positioning is captured. To cite this article: C. Lemaitre et al., C. R. Mecanique 334 (2006).     

Cristallisation par onde acoustique : le cas de l'héliumCrystallization by acoustic waves: the case of helium

We establish a theoretical model to explain the nucleation of a crystal of helium by an acoustic over-pressure. We explain the interfacial laws for this ultra-fast cristallization, close to the sound speed. Assuming spherical symmetry and taking into account the experimental data, we recover the dynamics of the growth and melting during an over-pressure impulse. To cite this article: M. Ben Amar et al., C. R. Mecanique 331 (2003).     

Structure initiale et propriétés de liquéfaction statique d'un sableInitial structure and static liquefaction properties of sand.

Based on the use of two different preparation procedures for reconstituting triaxial samples of sand, i.e. wet tamping and dry pluviation, significant differences in associated mechanical behaviour are observed on a reference sand with respect to the phenomenon of ‘static’ liquefaction. Wet tamping favours the initiation of liquefaction instability, whereas dry pluviation favours a more stable behaviour, less susceptible to liquefaction. Microscopic observation of corresponding sand specimens allows us to identify two well differentiated structures, i.e., for wet tamping, an irregular structure with predominance of aggregates (aggregated grains) and macropores, very contractant and unstable and, for dry pluviation, a more regular structure, without macropores, more dilatant and more stable. These observations show the importance of further characterization, based on the introduction of appropriate parameters, of the initial structure of sandy materials, strongly dependant upon their mode of formation (natural or artificial). To cite this article: N. Benahmed et al., C. R. Mecanique 332 (2004).     

Sources of Complexity in Human Systems

Complex is a special attribute we can give to many kinds of systems. Although it is used often as a synonym of difficult, it has a specific epistemological meaning, which is going to be shared by the incoming science of complexity. Difficult is an object which, by means of an adequate computational power, can be deterministically or stochastically predictable. On the contrary complex is an object which can not be predictable because of logical impossibility or because its predictability would require a computational power far beyond any physical feasibility, now and forever. For complexity refers to some observing system, it is always subjective, and thus it is defined as observed irreducible complexity. Human systems are affected by several sources of complexity, belonging to three classes, in order of descending restrictivity. Systems belonging to the first class are not predictable at all, those belonging to the second class are predictable only through an infinite computational capacity, and those belonging to the third class are predictable only through a trans-computational capacity. The first class has two sources of complexity: logical complexity, directly deriving from self-reference and Gödel's incompleteness theorems, and relational complexity, resulting in a sort of indeterminacy principle occurring in social systems. The second class has three sources of complexity: gnosiological complexity, which consists of the variety of possible perceptions; semiotic complexity, which represents the infinite possible interpretations of signs and facts; and chaotic complexity, which characterizes phenomena of nonlinear dynamic systems. The third class coincides with computational complexity, which basically coincides with the mathematical concept of intractability. Artificial, natural, biological and human systems are characterized by the influence of different sources of complexity, and the latter appear to be the most complex.