神经网络智能诊断技术在混凝土结构中的理论实现与展望

混凝土结构诊断技术是现代建筑工业中的一个重要课题。基于人工神经网络的智能诊断,其融合了计算机技术与神经网络技术,并结合了工程设备的特征,它具有非线性、快速、精确等优点,更接近工程实际;同时,具有美好的发展前景。     

浮选图像中泡沫边缘分割

矿物浮选过程中泡沫的表面特征是浮选性能的重要指标,它可以实时、直观地反映浮选效果的变化,准确分割泡沫的边缘信息是浮选过程中一项重要的任务。近年来,研究人员提出了各种浮选泡沫图像分割算法,但浮选泡沫图像中存在泡沫数量多、泡沫间粘连严重以及边缘不清晰等问题,现存的方法由于其特征提取能力有限,无法精确的分割泡沫边缘。据此,本文利用深度学习提出了一种基于多尺度融合的浮选泡沫图像边缘分割算法,该算法通过引入一种深度高分辨率的编码结构以及一种基于注意力的分层融合方法来增强模型的特征提取能力,从而提高对于浮选泡沫边缘的分割效果。具体而言,深度高分辨率的编码结构可以在不同分辨率层级上同时维护特征信息,使我们的网络模型可以有效地捕捉不同尺度的信息,在提高图像语义理解能力的同时能够保持更多的细节信息,提高处理高分辨率以及密集任务图像的能力。除此之外,本文设计了一种基于注意力的分层融合方法来充分融合深层和浅层的特征图,使融合得到的特征图趋向于更重要的特征信息,从而提高识别浮选泡沫的边界和精确定位浮选泡沫的能力。该算法在泡沫边界分割数据集上凭借58.25的泡沫IoU以及73.62的泡沫Fscore取得了最佳的性能,证明了我们提出的算法可以更加准确地分割浮选泡沫边缘。     

NUMERICAL SIMULATION OF THE FORMATION OF NANOPARTICLES IN AN IMPINGING TWIN-JET

The Large Eddy Simulation (LES) method has been used to solve the fluid momentum equations coupled with a convection-diffusion equation to study the formation of pollutant nanoparticles in a vehicular exhaust with impinging twin-jet. The functions of the space (S) between the two jets and the distance (H) from the exit of nozzle to the impingement plane are evaluated according to the distributions of pollutant nanoparticles. The results show that the nucleation produces a large number of nanoparticles, and gas-to-nanoparticle conversion mostly takes place in the interface region of the two jets, the circumambience of the jets, and region near the plane. The maximal particle size and maximal number concentration produced by both nucleation and coagulation appear around the region of free jet and the region near the plane, respectively. The significant differences for various spaces between the two jets are the number concentration and size distributions in the interface region of the two jets. For the case with larger space, more nanoparticles are produced by nucleation and coagulation. The more the distance from the exit of nozzle to the impingement plane, the lower is the number concentration and the fewer the particles distribute near the plane. Increasing the distance from nozzle to plane is beneficial to the reduction of nanoparticle formation.     

Research on relations between failure heights of overburden strata and ~] / mining face parameters and forecasting method

The commercial FEM software ANSYS was used to analyze the failure characteristics of overburden strata under the conditions of different lengths of mining faces. It was shown that the parameters of mining faces, especially the length was the important factor to the failure heights and shapes of overburden strata. Fuzzy mathematics and statistical methods were used to analyze the forecasting method of the failure height of overburden strata influenced by the parameters of mining face based on the measured data under the conditions of fully-mechanized mining of general hardness cover rocks. On the basis of these analyses, a new forecasting formula was gotten. The forecasting result conforms to the in situ measured value. The result has a very important application value in safe and high-efficient mining, and has a very important advancing function to theoretical studies.     

The creep experiment and theoretical model analysis of gas- containing coal

A creep experiment of preformed molding coal under different confining pressures were carried out using self-developed 3-triaxial creep loading device for gas-containing coal, which loaded by Shimadzu AGI-250 kN electrical servo-controlled stiffness testing machine. Based on the experimental results, the variation trend of axial deformation under different stress states was studied, and creep failure characteristics of gas-containing coal under different confining pressures were analyzed. The experimental results were identified with seven-component nonlinear viscoelasto-plastic creep model （Hohai model）, and the creep material parameters were obtained. The experimental result complies well with the theoretical value of this model. It indicates that creep constitutive relation of gas-containing coal can be expressed by nonlinear viscoelasto-plastic creep model correctly.     

基于大数据的颌骨囊性病变诊疗系统的研究

论文设计了基于c4.5和BP的颌骨囊性病变诊疗分析系统.对颌骨囊性病变的大数据进行分析处理.系统的分析结果可用于指导颌骨囊性病变的分类、诊断、治疗、预测、评估.实验表明,系统对大数据级别的颌骨囊性病变进行决策支持和特征诊断具有可行性和有效性.系统具有高吞吐能力、高扩展性、高容错性,有较好的推广使用价值.     

基于静态和动态混合分析的内存拷贝类函数识别

缓冲区溢出等内存错误漏洞的产生往往来自对内存拷贝类函数的不当使用. 对二进制程序中的内存拷贝类函数进行识别有利于发现内存错误漏洞. 目前针对二进制程序中内存拷贝类函数的识别方法主要借助静态分析来提取函数的特征、控制流、数据流等信息进行识别, 具有较高的误报率和漏报率. 为了提高对内存拷贝类函数识别的效果, 提出一种基于静态和动态混合分析的技术CPSeeker. 所提方法结合静态分析和动态分析各自的优势, 分阶段对函数的全局静态信息和局部执行信息进行搜集, 对提取到的信息进行融合分析, 进而识别二进制程序中的内存拷贝类函数. 实验结果表明, 尽管CPSeeker在运行时间上有所增加, 但在内存拷贝类函数识别的效果上, 其F1值达到了0.96, 远优于最新的工作BootStomp、SaTC、CPYFinder以及Gemini, 并且不受编译环境(编译器版本、编译器种类、编译器优化等级)的影响. 此外, CPSeeker在真实的固件测试中也有更好的表现.     

基于闭环矢量法的六足机器人关节位姿研究

设计出了一种新型的三自由度缩放式的并联机构,并把该机构作为六足步行机器人的腿机构.采用闭环矢量法求解腿机构运动位置的正运动学问题,用MATLAB软件编写函数来求解腿机构的最大腿行程和最大抬跨高度/研究表明,原动件较小的位移可以获得较大的足端运动空间.     

计算机网络工程实验优化探索

实验教学是计算机专业教学活动中的一个重要环节,现如今实验教学环节中存在着实验设备滞后、实验内容单一、学生实验积极性没有充分调动等众多问题。从以上三方面展开《计算机网络工程》课程的实验教学研究,提出将网络仿真工具应用于计算机网络实验教学、强调学生在实验中的主体作用、将实验项目模块化、搭建虚拟网络实验平台等多种优化策略。实践证明这些优化方案在网路工程实验环节中取得了良好效果。     

冶炼铜选尾渣对废石粗骨料充填料浆特性改性研究

针对废石粗骨料高浓度自流充填料浆离析、充填体分层严重,及冶炼铜渣污染大、处理难的矿山实际难题,本研究以铜渣尾砂作为细骨料对废石粗骨料充填料浆进行改性为指导思想,通过强度试验、塌落扩展度、分层度及泌水率试验研究了铜渣尾砂对废石高浓度充填的影响.结果表明,添加铜渣尾砂细骨料对充填体的早期强度几乎没有影响,但能大幅提高充填体...