基于PANDA软件的框架网坐标解算可靠性分析

应用高精度定位解算软件PANDA对框架网进行解算,并与Bernese解进行比较,探讨不同软件模型下框架网的点位坐标解算方法及其精度;通过算例表明PANDA软件的定位解算结果与Bernese的解有较好的一致性,表明PANDA软件有较高的可靠性和稳定性。     

基于CAN总线电阻的最大长度和节点数求解方法

以82C250作为CAN控制器与物理层总线之间的接口,通过CAN控制器向总线差动发送与接收,建立CAN总线网络分布等效电路,推导出最大总线长度和节点数,从而为选择CAN总线来组建船舶监控系统提供理论依据,解决了限制CAN总线速度进一步提高的缺陷,并能迅速发现和排除运行中的故障,增加安全可靠性.     

Lift distribution based design of two-dimensional sections

A lift distribution based section design method has been proposed. Through Newton-Raphson iterations, the section geometry is efficiently designed to meet the requirements for total lift and lift distribution. The effect of fluid viscosity on total lift is taken into account by coupling Reynolds-averaged Navier-Stokes (RANS) simulation with the potential-flow based design procedure. The present method avoids the difficulty of assigning velocity or pressure distributions on section surfaces. As the loading and thickness distributions are expressed in parametric forms, it is easy to ensure that the designed geometry is continuous and smooth. The effects of lift and thickness distributions on cavitation bucket are numerically investigated. A shift of lift loading towards the aft part of section tends to decrease the margin of back cavitation, while the width of cavitation bucket can be kept almost unchanged. To have a wider cavitation bucket, one can increase the leading edge radius, move properly the location of maximum thickness towards the leading edge, or decrease the curvature at the location of maximum thickness.     

湍流入流下泵喷推进器力谱特性研究

[目的]旨在解决泵喷推进器在湍流入流作用下非定常力谱不明的问题.[方法]通过湍流生成栅格和频谱合成法相结合的方法,产生具有时间脉动和空间相干结构的湍流,然后将此种方法与模拟相结合,获得泵喷推进器非定常力的宽带谱.[结果]结果显示,所采用的计算方法能够获得泵喷推进器导管、定子和转子合理的非定常力宽带谱;泵喷推进器非定常力...     

基于生成对抗网络的隧道裂缝自动分割算法研究

传统裂缝识别技术需要大量带标签的裂缝图像作为试验数据集，为减少裂缝图像标注的工作量，利用生成对抗网络图像分割的特性与优势，构建一种Crack-GAN网络用于隧道裂缝自动分割。Crack-GAN网络结构集成2个模块：融合残差化U-Net网络的生成器网络和利用全卷积网络生成置信图的判别器网络。首先U-Net模块使用密集的残差模块来生成保留细粒度信息的深层表示，然后判别器来判断输入真假，并以端到端的方式训练，再经过生成对抗模型之间不断迭代，使生成模型达到分割裂缝的最优状态。试验表明，Crack-GAN网络在自制铁路隧道裂缝数据集上的像素准确性为98.35%,精准率为71.23%,召回率为80.78%,F1得分为75.98%,单次检测时间80 ms,综合表现优于U-Net和CrackSegNet。