Artificial Neural Network and Fuzzy Logic Based Techniques for Numerical Modeling and Prediction of Aluminum-5%Magnesium Alloy Doped with REM Neodymium

In this study, the mechanical properties of aluminum-5%magnesium doped with rare earth metal neodymium were evaluated. Fuzzy logic (FL) and artificial neural network (ANN) were used to model the mechanical properties of aluminum-5%magnesium (0-0.9 wt%) neodymium. The single input (SI) to the fuzzy logic and artificial neural network models was the percentage weight of neodymium, while the multiple outputs (MO) were average grain size, ultimate tensile strength, yield strength elongation and hardness. The fuzzy logic-based model showed more accurate prediction than the artificial neutral network-based model in terms of the correlation coefficient values (R).

Modeling of grain size in isothermal compression of Ti-6Al-4V alloy using fuzzy neural network

Isothermal compression of Ti-6Al-4V alloy was conducted in the deformation temperature range of 1093-1303 K, the strain rates of 0.001, 0.01, 0.1, 1.0, and 10.0 s-1, and the height reductions of 20%-60% with an interval of 10%. After compression, the effect of the processing parameters including deformation temperature, strain rate, and height reduction on the flow stress and the microstructure was investigated. The grain size of primary a phase was measured using an OLYMPUS PMG3 microscope with the quantitative metallography SISC IAS V8.0 image analysis software. A model of grain size in isothermal compression of Ti-6A1-4V alloy was developed using fuzzy neural net- work （FNN） with back-propagation （BP） learning algorithm. The maximum difference and the average difference between the predicted and the experimental grain sizes of primary a phase are 13.31% and 7.62% for the sampled data, and 16.48% and 6.97% for the non-sampled data, respectively. It can be concluded that the present model with high prediction precision can be used to predict the grain size in isothermal compression of Ti-6Al-4V alloy.

Optimal Packet Size of Underwater Sensor Networks

Parameter optimization of nodes communication is the foundation of underwater sensor networks.The packet size is an important indicator of the impact of communication performance.As a result,the optimal packet size selection is a critical issue in improving the communication performance.This paper aims to make a model reflecting the communication characteristics as the optimization target,because underwater sensor networks have the characteristics of high time delay,high energy consumption and high bit error rate.Finally,simulation experiments and theory have demonstrated the effectiveness and timeliness of simultaneous perturbation stochastic approximation(SPSA) algorithm.

Quality of Service Analysis of Ethernet Network Based on Packet Size

Ethernet network, standardized by IEEE 802.3, is vastly installed in Local Area Network (LAN) for cheaper cost and reliability. With the emergence of cost effective and enhanced user experience needs, the Quality of Service (QoS) of the underlying Ethernet network has become a major issue. A network must provide predictable, reliable and guaranteed services. The required QoS on the network is achieved through managing the end-to-end delay, throughput, jitter, transmission rate and many other network performance parameters. The paper investigates QoS parameters based on packet size to analyze the network performance. Segmentation in packet size larger than 1500 bytes, Maximum Transmission Unit (MTU) of Ethernet, is used to divide the large data into small packets. A simulation process under Riverbed modeler 17.5 initiates several scenarios of the Ethernet network to depict the QoS metrics in the Ethernet topology. For analyzing the result from the simulation process, varying sized packets are considered. Hence, the network performance results in distinct throughput, end-to-end delay, packet loss ratio, bit error rate etc. for varying packet sizes.

Growing actin networks regulated by obstacle size and shape

Growing actin networks provide the driving force for the motility of cells and intracellular pathogens. Based on the molecular-level processes of actin polymerization, branching, capping, and depolymerization, we have developed a modeling framework to simulate the stochastic and cooperative behaviors of growing actin networks in propelling obstacles, with an emphasis on the size and shape effects on work capacity and filament orientation in the growing process. Our results show that the characteristic size of obstacles changes the protrusion power per unit length, without influencing the orientation distribution of actin filaments in growing networks. In contrast, the geometry of obstacles has a profound effect on filament patterning, which influences the orientation of filaments differently when the drag coefficient of environment is small, intermediate, or large. We also discuss the role of various parameters, such as the aspect ratio of obstacles, branching rate, and capping rate, in affecting the protrusion power of network growth.

Major impact of queue-rule choice on the performance of dynamic networks with limited buffer size

We investigate the similarities and differences among three queue rules,the first-in-first-out(FIFO)rule,last-in-firstout(LIFO)rule and random-in-random-out(RIRO)rule,on dynamical networks with limited buffer size.In our network model,nodes move at each time step.Packets are transmitted by an adaptive routing strategy,combining Euclidean distance and node load by a tunable parameter.Because of this routing strategy,at the initial stage of increasing buffer size,the network density will increase,and the packet loss rate will decrease.Packet loss and traffic congestion occur by these three rules,but nodes keep unblocked and lose no packet in a larger buffer size range on the RIRO rule networks.If packets are lost and traffic congestion occurs,different dynamic characteristics are shown by these three queue rules.Moreover,a phenomenon similar to Braess’paradox is also found by the LIFO rule and the RIRO rule.

An Artificial Neural Network for Estimating Sizes of Spot Welding Nuggets

Sizes of nuggets are often used to evaluate spot weld quality in production. This paper presents a neural estimator used to carry out non-destructive on-line analysis of spot weld quality in which trained ANN functions to map dynamic resistance characteristics into sizes of spot weld nuggets and results confirm the validity of neural network for this type of application.

A NEURAL NETWORK-BASED MODEL FOR PREDICTION OF HOT-ROLLED AUSTENITE GRAIN SIZE AND FLOW STRESS IN MICROALLOY STEEL

For the great significance of the prediction of control parameters selected for hot-rolling and the evaluation of hot-rolling quality for the analysis of prod uction problems and production management, the selection of hot-rolling control parameters was studied for microalloy steel by following the neural network principle. An experimental scheme was first worked out for acquisition of sample data, in which a gleeble-1500 thermal simolator was used to obtain rolling temperature, strain, stain rate, and stress-strain curves. And consequently the aust enite grain sizes was obtained through microscopic observation. The experimental data was then processed through regression. By using the training network of BP algorithm, the mapping relationship between the hotrooling control parameters (rolling temperature, stain, and strain rate) and the microstructural paramete rs (austenite grain in size and flow stress) of microalloy steel was function appro ached for the establishment of a neural network-based model of the austeuite grain size and flow stress of microalloy steel. From the results of estimation made with the neural network based model, the hot-rolling control parameters can be effectively predicted.

融合GBWO与ENN的人体尺寸预测模型

为了提高人体尺寸预测的效率和准确性,该文提出了GBWO-ENN(Grey Black Wolf Optimization-Elman Neural Network)的方法。针对传统灰狼算法易于陷入局部最优和无法平衡全局与局部搜索的平衡性问题,提出了GBWO算法。该算法融合黑寡妇优化算法中蜘蛛的运动方式对灰狼优化算法中α狼位置更新进行了优化,通过非线性递减的方法降低了收敛系数,并且提出了按位置等级更新种群的策略。随后采用GBWO算法对Elman神经网络的权值和阈值进行优化,并将GBWO-ENN模型应用于三维人体尺寸预测。实验结果表明,GBWO-ENN模型结构简单,能够准确预测人体尺寸,具有较好的预测能力。

基于YOLOv3的金属表面缺陷检测研究

为了解决金属表面缺陷检测的漏检、误检等问题,提出了一种改进YOLOv3算法。首先,使用动态激活函数替换主干特征提取网络中所有残差块的激活函数,并加入了混合注意力机制,强化其对复杂缺陷目标的特征提取能力。然后,在特征金字塔网络部分新增一个104×104的特征层,并将浅层网络与深层网络进行逐层特征融合,增强算法对小缺陷目标检测的敏感性。最后,利用K-Means++聚类算法替换K-Means聚类算法,筛选出适用于金属表面缺陷检测的最优先验框尺寸,使目标定位更加准确。实验结果表明,改进YOLOv3算法的每秒检测帧数(frames per second,FPS)可达到32.3,平均精度均值(mean average precision,mAP)可达到78.69%,检测性能得到了明显提升。

网络社区用户知识分享行为的机理研究——基于网络规模效应的调节作用

网络社区已经成为知识传播的重要方式。用户数量规模以及知识分享活跃度对网络社区的发展尤为重要。基于社会交互关系、信任和共同语言3个维度,从网络规模效应调节作用的视角构建了用户知识分享行为的机理模型。通过问卷调查获取数据对理论模型进行实证分析,研究表明社会交互关系、信任和共同语言对用户的知识分享意愿有显著正向影响;同边网络效应在信任对用户知识分享意愿的影响中起负向调节作用;跨边网络效应在社会交互关系、共同语言对用户知识分享意愿的影响中起正向调节作用。最后,从构建良好的知识网络社区环境、加大知识网络社区的品牌宣传、建立科学的知识分享激励机制等方面提出了具体对策。

基于长短时记忆网络的山区中小流域降雨径流模拟

洪水预报是流域防洪减灾的重要非工程措施之一。目前我国中小河流暴雨洪水灾害频发,但应对短历时强降雨的洪水预报能力仍不强。以安徽省东部山区中小流域为研究对象,引入长短时记忆网络建立流域降雨径流模型,探讨其在山区中小流域的洪水模拟效果。结果表明,考虑降雨输入的空间差异可提升深度学习模型降雨径流模拟预测性能,且长短时记忆网络能够取得优于传统人工神经网络的精度;长短时记忆网络模型有效建立了流域降雨与径流间的复杂非线性关系,模型在所选流域内场次洪水的峰值模拟效果较好,训练、测试集场次洪水峰值合格率均在90%以上;长短时记忆网络内部结构特征与流域水文过程具有较好的相似性,对山区中小流域暴雨洪水非线性关系拟合效果突出。

多源数据视角下黄河流域城市体系的规模等级与网络结构分析

基于宏观统计、微观企业和交通大数据,从规模和网络2个方面及“城市节点-发展轴线-空间格局”3个维度,分析黄河流域城市体系的规模等级和网络结构特征以及二者关系。结果表明:黄河流域城市体系呈现多中心分布特征,规模等级趋于均衡,形成了以兰州-西安-郑州-济南-青岛为主导的发展轴线和“Ψ”型空间格局。规模等级和网络结构成拉平的“S型”曲线关系,可划分为低水平低速耦合、中水平高速耦合和高水平低速耦合3个阶段。

带襟翼导轨翼肋后缘尺寸-拓扑综合优化的摄动神经网络代理模型法

带襟翼导轨的翼肋后缘设计需要确定肋缘条、腹板的尺寸和肋腹板的拓扑形状,对此提出了一种针对尺寸-拓扑综合优化的摄动神经网络(perturbation neural network, PNN)代理模型法.其基本思想是基于拓扑优化对参数的敏感性,引入了对试验设计(design of experiments, DOE)样本点的摄动,通过过滤手段捕获拓扑突变点,并降低数值噪声,极大地提高了代理模型的预测精度,将拓扑优化过程作为黑盒,直接建立起尺寸变量与拓扑优化后结构响应的代理模型.最后在代理模型上进行优化,得到了结构尺寸与拓扑形状的最优组合.该文完成了一个翼肋后缘优化典型算例,证明了该方法的有效性和优越性.

基于三维裂隙网络的岩体剪切特性尺寸效应分析

为研究复杂裂隙岩体剪切特性的尺寸效应,以西藏怒江松塔水电站PDC3平硐为例,在现场采集裂隙信息的基础上,基于颗粒流PFC3D软件通过统计学方法及蒙特卡洛原理生成三维裂隙网络,构建等效岩体,进行大尺度直剪模拟试验得到裂隙岩体的抗剪强度、内摩擦系数及内聚力,分别研究其尺寸效应并确定不同参数下的REV尺寸,继而对相关参数和试样尺寸分别进行非线性回归拟合,探究其函数关系。结果表明:抗剪强度、内聚力及内摩擦系数随着试样尺寸的变化而趋于稳定,且不同参数得到的REV大小不同;综合分析不同参数的REV尺寸,得出研究区裂隙岩体的剪切特性REV尺寸为11 m×11 m×11 m;通过函数关系拟合,抗剪强度、内聚力和内摩擦系数与试样尺寸之间均存在指数函数关系。该成果为研究复杂裂隙岩体的力学REV以及合理确定力学参数提供了依据。

基于粒子群优化算法的电弧增材制造焊道尺寸反向传播神经网络预测模型

选取焊接电流、送丝速度、焊接速度及基板温度作为输入变量,焊道熔宽和余高作为输出变量,选择粒子群优化(PSO)算法中的最优粒子惯性权重和学习因子,构建熔化极惰性气体保护电弧增材制造316L不锈钢PSO反向传播(PSO-BP)神经网络模型。结果表明:PSO-BP神经网络模型预测的焊道熔宽与期望值的均方根误差、最大相对误差与平均相对误差分别为0.386,13.477%,2.580%,焊道余高的分别为0.152,10.372%,2.810%;相较于BP神经网络模型,PSOBP神经网络模型对焊道尺寸的预测精度更高,稳定性更强。

高铁网络对异质性城市创新的机制研究

高铁网络的时空压缩效应对创新要素流动和创新能力空间格局变化产生了重要影响,但不同规模城市的异质性特征对创新要素的筛选效应并没有引起重视。基于新经济地理异质性假设,采用2008—2019年的中国城市面板数据,基于连续型DID估计方法研究高铁网络对城市创新发展的影响。结果表明,高铁网络显著促进了城市创新能力提升,但这种促进作用需要在城市规模达到一定门槛时才能显现,且这种异质性城市创新效应在城市的创新禀赋上也得以体现,创新能力越强的城市越易于将高铁网络传送的创新要素快速转化为创新产出。进一步分析发现,在城市人口规模达到一定门槛值后,城市才能基于高铁网络改善的连通性从周边中小城市获得人才集聚、金融资本集聚和产业集聚的红利效应,实现城市创新发展的目标。未来的城市发展需要进一步优化现有创新激励政策,将促进创新的路径从主要依靠资本投入向人才和产业吸引转移,充分发挥城市对创新要素集聚的重要作用,通过客观准确地进行差异化城市功能定位加强产业协作,扩大创新空间溢出效应,以点带面全面提升城市创新。

基于数据到达间隔的网络大流检测方法

大流检测是网络监管中的重要任务。现有的检测方法主要围绕流的大小来进行筛选评判。然而,真实的大流往往与数量众多的老鼠流混合在一起,尤其是当路由器存储空间有限时,依靠所能记录下来的流大小信息通常不能准确实现大流检测。为此提出一种新的大流检测方法,该方法根据流的新旧及数据到达间隔来滤除老鼠流。实验显示该方法在有限的存储空间条件下,表现出良好的检测准确性。

中国城市“双碳”网络关注度的时空差异及影响因素

基于百度指数平台采集2012—2021年中国287个地级市的“双碳”相关网络搜索指数,运用赫芬达尔指数、多尺度地理加权回归等方法分析“双碳”网络关注度的时空差异和影响因素。结果表明:(1)时间上,中国城市“双碳”网络关注度在波动中呈上升态势,区域关注差异先增大后减小至最低水平;空间上,各城市之间网络关注度分布的不均衡性较强,总体呈现出“东—中—西”依次降低态势。(2)省域网络关注度差异程度呈现波动中不断扩大的趋势,空间集聚性不显著;六大区域关注度差异始终处于高位且呈波动上升态势,空间集聚程度呈现出平稳波动态势。(3)影响网络关注度的因素存在空间异质性。人均GDP和常住人口的正向影响效应显著,规模以上工业企业数量的负向影响呈现出南北差异格局,工业氮氧化物排放量的空间作用效应相对均衡,政策因子得分的作用效应呈现显著的东西差异。

西北江三角洲河网浮游植物叶绿素a粒径特征研究

为探索浮游植物叶绿素a粒径的时空变化及其影响因素,调查研究了西北江三角洲河网水域2021年枯水期(2月份和12月份)和丰水期(5月份和8月份)的水环境和浮游植物叶绿素a粒径特征。结果显示,叶绿素a总浓度呈现明显的季节差异,最高值出现在8月份,最低值出现在5月,分析原因主要与径流量的大小、水体pH和氧化还原电位的变化有关。叶绿素a总浓度的空间特征显示,广州周边站位的叶绿素a浓度明显偏高,主要与营养盐的空间分布格局和水体的交换能力有关。粒径特征的结果显示,小型和微型浮游植物是珠三角河网水体叶绿素a的主要贡献者,呈现交替优势。季节特征显示,2-5月份微型叶绿素a占主导地位,8月份和12月份小型叶绿素a占主导地位,分析原因主要与径流量的大小、水体pH和光照有关。空间特征显示,小型叶绿素a在8月份和12月份占比的最高值往往出现在河网中部的站位,微型叶绿素a占比的最高值往往出现在远离广州市的站位点,分析原因主要与溶解氧和硝酸盐含量有关。冗余分析结果表明pH、DO、水下光度、TP和硝酸盐含量是影响浮游植物粒径分布的主要影响因子。综上,物理因子包括径流量的大小和水体pH的变化影响叶绿素a粒径特征的季节变动,而化学因素如营养盐含量影响叶绿素a粒径特征的空间分布。