Long-term oxidation behaviour of Ni3Al alloys with and without chromium additions

The oxidation behaviour of Ni₃Al alloys with and without chromium additions was studied after long-term exposure in an air furnace over a wide temperature range from 560–1300C. The chromium-containing alloy exhibited a better oxidation resistance below 1150C. Above this temperature the penetration depth of oxides into and along the grain boundaries in the chromium-containing alloy became larger than that in the alloy without chromium. Also the penetration depth increased drastically at 1150C for both alloys. Through energy dispersive X-ray spectroscopy, X-ray diffraction and X-ray photoelectron spectroscopy analyses, the compositions and structures of the oxide scales for both alloys were also identified.     

快速卷积算法的综述研究

卷积神经网络是深度学习算法应用最广泛的方向之一,目前卷积神经网络的应用不仅仅是停留在科技领域,已经扩展到医学、军事等领域,并且已在相关领域发挥着巨大的作用。卷积是卷积神经网络中最为核心的一部分,卷积运算占整个网络70%以上的时间,所以针对卷积运算的加速研究就显得十分重要。首先介绍近年来的卷积算法,并对其复杂度进行分析,总结了这些算法各自的优点和不足,最后对其理论研究和应用领域可能存在的突破进行了探讨和展望。     

基于双重权重偏差建模的无监督域适应

无监督域适应(Unsupervised Domain Adaptation,UDA)是一类新兴的机器学习范式,其通过对源域知识在无标记目标域上的迁移利用,来促进目标域模型的训练。为建模源域与目标域之间的域分布差异,最大均值差异(Maximum Mean Discrepancy,MMD)建模被广泛应用,其对UDA的性能提升起到了有效的促进作用。然而,这些方法通常忽视了领域之间对应类规模与类分布等结构信息,因为目标域与源域的数据类规模与数据分布通常并非一致。为此,文中提出了一种基于跨域类和数据样本双重加权的无监督域适应模型(Sample weighted and Class weighted based Unsupervised Domain Adaptation Network,SCUDAN)。具体而言,一方面,通过源域类层面的适应性加权来调整源域类权重,以实现源域与目标域之间的类分布对齐;另一方面,通过目标域样本层面的适应性加权来调整目标域样本权重,以实现目标域与源域类中心的对齐。此外,文中还提出了一种CEM(Classification Expectation Maximization)优化算法,以实现对SCUDAN的优化求解。最后,通过对比实验和分析,验证了所提模型和算法的有效性。     

基于图卷积神经网络的药物靶标作用关系预测方法

药物-靶标作用关系预测在药物研发以及药物重定位中扮演着重要角色,但现有的机器学习方法在正负样本高度不平衡的数据上仍存在预测能力不足的问题.为此,提出一种基于图卷积神经网络的药物靶标作用关系预测方法.该方法首先构造一个结合多种药物(靶标)相关信息的异质信息网络,然后采用图卷积神经网络在此异质信息网络上学习得到能精确表达每个节点拓扑特征及邻居特征信息的低维向量表征,最后利用这些向量信息通过向量空间投影预测节点间概率的评分.在DrugBank_FDA和Yammanishi_08数据集上进行的药物-靶标作用关系预测的对比实验中,所提方法的AUPR(Area Under the Precision-Recall Curve)值都优于其他4种方法,并且在较大型数据集上也有较好的表现.实验结果表明,所提方法提高了样本高度不平衡时的药物-靶标作用关系预测性能;同时在生物药物数据库上的实验也验证了所提方法所发现的未知药物-靶标作用关系的有效性.     

基于SSC-BP神经网络的异常检测算法

针对物联网环境下产生的新型网络攻击的数量持续上升和复杂性不断升高,传统的异常检测算法误报率高、检测率低以及数据量大而造成计算困难等问题,提出了一种基于子空间聚类(Subspace Clustering,SSC)和BP神经网络相结合的异常检测算法.首先在网络数据集上通过子空间聚类算法中最常用的CLIQUE算法得到不同的子空间;其次对不同子空间中的数据进行BP神经网络异常检测,计算预测误差值,通过与预先设定好的精度进行比较,来不断更新阈值进行修正,以达到提高识别网络攻击的能力.仿真实验采用NSL-KDD公开数据集和物联网环境下的网络攻击数据集,将NSL-KDD公开数据集分割为4种单一攻击子集和1种混合攻击子集,通过与K-means,DBSCAN,SSC-EA以及k-KNN异常检测模型进行比较,在混合攻击子集中,SSC-BP神经网络模型的检测率比传统的K-means模型的检测率提高了6％,误检率降低了0.2％;而在4种单一攻击子集中,SSC-BP神经网络模型都能以最低的误检率检测出最多的受到攻击的网络.在物联网环境下的网络攻击数据集上,SSC-BP神经网络模型的性能均优于其他几种对比模型.     

Mixed H2/H∞ State Feedback Attitude Control of Microsatellite Based on Extended LMI Method

For the appearance of the additive perturbation of controller gain when the controller parameter has minute adjustment at the initial running stage of system,to avoid the adverse effects,this paper investigates the mixed H_2/H_∞ state feedback attitude control problem of microsatellite based on extended LMI method.Firstly,the microsatellite attitude control system is established and transformed into corresponding state space form.Then,without the equivalence restriction of the two Lyapunov variables of H_2 and H∞performance,this paper introduces additional variables to design the mixed H_2/H_∞ control method based on LMI which can also reduce the conservatives.Finally,numerical simulations are analyzed to show that the proposed method can make the satellite stable within 20 s whether there is additive perturbation of the controller gain or not.The comparative analysis of the simulation results between extended LMI method and traditional LMI method also demonstrates the effectiveness and feasibility of the proposed method in this paper.     

空间遥感器反射镜组件的设计与有限元分析

针对某空间遥感器反射镜的设计指标要求,分析并优化了反射镜的支撑形式和结构参数,得到了质量为23.7kg,轻量化率达到76.2%的反射镜结构。在反射镜基本构型确定的基础上,设计了镜体的支撑结构,通过合理设计柔性卸载结构满足了反射镜结构系统的动静态刚度和热尺寸稳定性要求。采用有限元软件ABAQUS对反射镜组件进行了精确有限元分析,分析结果表明,在1g重力作用下反射镜的面形精度RMS达到3.66nm,在4℃均匀温升载荷作用下反射镜的面形精度RMS达到4.16nm,在1g重力和4℃均匀温升载荷耦合作用下反射镜的面形精度RMS达到5.51nm,反射镜组件的一阶固有频率为137.51Hz。得到的结果显示该反射镜组件完全满足设计指标要求。     

采用多分辨率广义S变换的电能质量扰动识别

为提高电能质量复合扰动识别能力,提出一种采用多分辨率广义S变换(multiresolution generalized S-transform,GST)的扰动识别方法.首先,将信号频谱分为低频、中频、高频3个频域,分别设定窗宽调整因子,使其在各个频域具有不同的时-频分辨率,满足不同扰动信号识别要求.并针对高频振荡识别问题,设计基于基频傅里叶谱特征的自适应窗宽调整方法.在此基础上,提取6种特征用于构建决策树.最后,提出最小分类损失原则,确定决策树节点分类阈值,设计扰动分类器.仿真与实测信号实验证明,新方法能够准确识别含5种复合扰动在内的13种扰动.相较于S变换、广义S变换和Hyperbolic S变换,新方法具有更好的特征表现能力,分类效果好,抗噪声干扰能力强.     

An effective thermal conductivity model for unsaturated compacted bentonites with consideration of bimodal shape of pore size distribution

An effective thermal conductivity model was proposed for unsaturated compacted bentonites with consideration of the bimodal shape of pore size distribution curves.The pores of soils were grouped into two dominant pore size modes corresponding to the intra-and inter-particle pores,and were simulated with randomly distributed spheroidal inclusions of different aspect ratios.With the assumption of preferential invasion of the wetting fluid(water)into pores of smaller sizes and by virtue of the analytical solution to the inhomogeneous inclusion problem in heat conduction,the model was developed using the Mori-Tanaka(MT),Ponte Casta?eda-Willis(PCW)and self-consistent(SC)homogenization approaches for different considerations of the interactions between pores and the solid phase.The proposed model is functions of the thermal conductivities of the solid,liquid and gas phases,porosity,the degree of saturation,the aspect ratios of pores and/or soil particles,and the fraction of the smaller group of pores.The proposed model was validated against five sets of laboratory measurement data on the MX-80,FEBEX,Kunigel-V1 and GMZ01 bentonites,showing a good agreement between the model predictions and the laboratory measurements.The responses of the model with respect to the geometries of pores and solid particles were examined.Compared to series-parallel structural models,the proposed model may overall exhibit better performance if proper homogenization schemes are adopted,but as an advantage,the model has clearer physical mechanisms and a smaller number of parameters.     

超声波辅助电镀沉积Ni镀膜工艺研究

采用超声波辅助电镀沉积工艺方法研究工艺参数对电镀沉积金属镍镀膜的表面形貌及力学性能的影响规律.实验结果表明:超声波辅助电镀沉积镍镀层时,超声波能够细化镀层的晶粒,降低镀层的孔隙率,同时显著地改善异性工件电镀层的均匀性.当超声波作用时间20 min时,电镀镍镀层的孔隙率为2个/cm2.随着超声波作用时间的延长,电镀镍镀层的显微硬度呈现出先增加后降低的变化趋势.