北京交通拥堵指数的多标度扩散熵分析

提供了一种基于标度指数谱的多标度扩散熵分析方法（MDEA）描述北京交通分形动态系统。通过扩散熵分析（DEA）与滑动拟合窗口方法相结合,分析2011年北京交通拥堵指数,展现标度性质的更多信息,提供更可靠的分析。模拟结果表明,在较大的标度范围内,标度指数呈现较大波动,且具有它们自身的指数模式。多标度扩散熵分析方法对于度量交通拥堵指数以及对不同交通状态进行分类提供了一种新的方法。     

CHARACTERIZATION OF VELOCITY SINGULARITY AND SUPPRESSION OF NOISE

ＣＨＡＲＡＣＴＥＲＩＺＡＴＩＯＮＯＦＶＥＬＯＣＩＴＹＳＩＮＧＵＬＡＲＩＴＹＡＮＤＳＵＰＰＲＥＳＳＩＯＮＯＦＮＯＩＳＥ￥Ｓｏｎｇ，ＳｈｏｕｇｅｎＴａｎｇ，Ｊｉｎｔｉａｎ（ＤｅｐａｒｔｍｅｎｔｏｆＧｅｏｌｏｇｙ，ＣｅｎｔｒａｌＳｏｕｔｈＵｎｉｖｅｒｓｉｔ...     

Application of multiscale modeling in the coating formation simulation of APS PYSZ TBCs

The process parameters of atmospheric plasma spraying (APS) influence the coating formation and properties of partially yttria stabilized zirconia (PYSZ) thermal barrier coatings (TBC). Simulations can be used to investigate this dependency and to design the coating process for a targeted production of TBCs. A whole process simulation was realized by modeling the linked subprocesses: plasma torch, plasma free jet, powder particles characteristics, and coating formation. The coating formation can be described by model approaches with different physical assumptions and geometric scales. One approach is the simulation of single powder particles hitting the substrate surface. An alternative macroscale finite element model (FEM) model approach is applied in the coating formation simulation. A group of particles is pooled in a splash that is dependent on the precalculated particle distribution in front of the substrate. A third modeling approach is applied to calculate effective mechanical and thermodynamical properties of coatings dependent on the experimentally obtained or calculated microstructure of the PYSZ TBC, which is based on different homogenization methods. The application of three simulation approaches in the whole process simulation of APS is discussed; advantages and disadvantages are elucidated. Results based on simulation and experiments are presented for a variation of process parameters. Missing links in the multiscale approach are detected to make suggestions for future modeling and simulation work. This article was originally published inBuilding on 100 Years of Success, Proceedings of the 2006 International Thermal Spray Conference (Seattle, WA), May 15–18, 2006, B.R. Marple, M.M. Hyland, Y.-Ch. Lau, R.S. Lima, and J. Voyer, Ed., ASM International, Materials Park, OH, 2006.     

基于多尺度分割的图像识别残差网络研究

深度卷积神经网络能够解决复杂的计算机视觉问题,被广泛应用于图像识别任务中。在基于深度卷积神经网络的图像识别过程中,增加网络的深度和宽度能够产生丰富的特征信息,使用多尺度分割方法能够有效减少冗余的特征信息。然而,增加网络的深度和进行多尺度分割都会影响识别速度。如何在保证精度的同时提高识别速度,成为设计高效网络的关键问题。通过增加网络宽度的方法对ResNet残差网络进行改进,在保证精度的基础上提升识别速度。使用ResNet-D中的残差结构并减少网络长度,得到长度只有7层的残差网络,同时对HS-ResNet中的多尺度分割方法进行优化,只保留最后一次连接合并操作,得到图像识别残差网络SSRNet。在CIFAR 10和CIFAR 100数据集上的实验结果显示,SSRNet速度最高较ResNet网络提升7倍多,同时错误率最高下降8.81%,表明缩短网络长度可大幅加快图像识别速度,同时结合多尺度分割方法能够有效提升识别精度。     

Multilevel assessment method for reliable design of composite structures

This work aims at demonstrating the interest of a new methodology for the design and optimization of composite materials and structures. Coupling reliability methods and homogenization techniques allow the consideration of probabilistic design variables at different scales. The main advantage of such an original micromechanics-based approach is to extend the scope of solutions for engineering composite materials to reach or to respect a given reliability level. This approach is illustrated on a civil engineering case including reinforced fiber composites. Modifications of microstructural components properties, manufacturing process, and geometry are investigated to provide new alternatives for design and guidelines for quality control.     

A two‐scale generalized finite element approach for modeling localized thermoplasticity

Predicting localized, nonlinear, thermoplastic behavior and residual stresses and deformations in structures subjected to intense heating is a prevalent challenge in a range of modern engineering applications. The authors present a generalized finite element method targeted at this class of problems, involving the solution of intrinsically parallelizable local boundary value problems to capture localized, time‐dependent thermo‐elasto‐plastic behavior, which is embedded in the coarse, structural‐scale approximation via enrichment functions. The method accommodates approximation spaces that evolve in between time or load steps while maintaining a fixed global mesh, which avoids the need to map solutions and state variables on changing meshes typical of traditional adaptive approaches. Representative three‐dimensional examples exhibiting localized, transient, nonlinear thermal and thermomechanical effects are presented to demonstrate the advantages of the method with respect to available approaches, especially in terms of its flexibility and potential for realistic future applications in this area. Parallelism of the approach is also discussed. Copyright © 2016 John Wiley & Sons, Ltd.     

木材内部水分扩散机理及多尺度模型研究概述

木材在纤维饱和点(FSP)以下干燥时,内部水分主要是以扩散的形式迁移;水分扩散的主要动力是含水率梯度和温度梯度。不同尺度扩散模型的建立可以更好地描述干燥过程中水分的动态迁移。从水分扩散形式、扩散机理和不同尺度的扩散模型3个方面总结了木材干燥过程中水分扩散的研究进展,提出模型研究中存在的问题及相应措施,为扩散模型的建立和有效应用提供参考。     

Classical and variational multiscale LES of the flow around a circular cylinder on unstructured grids

The effects of numerical viscosity, subgrid scale (SGS) viscosity and grid resolution are investigated in LES and VMS-LES simulations of the flow around a circular cylinder at Re=3900 on unstructured grids. The separation between the largest and the smallest resolved scales in the VMS formulation is obtained through a variational projection operator and finite-volume cell agglomeration. Three different non-dynamic eddy-viscosity SGS models are used both in classical and in VMS-LES. The so-called small-small formulation is used in VMS-LES, i.e. the SGS viscosity is computed as a function of the smallest resolved scales. Two different grid resolutions are considered. It is found that, for each considered SGS model, the amount of SGS viscosity introduced in the VMS-LES formulation is significantly lower than in classical LES. This, together with the fact that in the VMS formulation the SGS viscosity only acts on the smallest resolved scales, has a strong impact on the results. However, a significant sensitivity of the results to the considered SGS model remains also in the VMS-LES formulation. Moreover, passing from classical LES to VMS-LES does not systematically lead to an improvement of the quality of the numerical predictions.     

基于NSCT和泰勒级数的图像超分辨率重建

结合非下采样Contourlet变换(NSCT)的多尺度、方向特性、各向异性、平移不变性以及泰勒级数高效逼近的优点,提出一种利用泰勒级数插值代替双线性插值的超分辨率图像重建算法.实验结果显示,该算法可以较好地恢复图像的细节信息和纹理特征,有效抵抗高斯噪声的干扰.