变电缺陷运检是智慧变电站的重要组成部分,其可以智能化识别定位变电缺陷,为变电维修提供依据。使用现行方法运检效果不佳,不仅缺陷漏检比例比较高,而且缺陷定位相对偏差比较大,因此本文提出基于地理信息系统(Geographic Information Sy...变电缺陷运检是智慧变电站的重要组成部分,其可以智能化识别定位变电缺陷,为变电维修提供依据。使用现行方法运检效果不佳,不仅缺陷漏检比例比较高,而且缺陷定位相对偏差比较大,因此本文提出基于地理信息系统(Geographic Information System,GIS)融合的智能变电缺陷运检方法。利用无人机搭载工业相机和GIS设备,感知变电图像和地理信息数据,对图像进行均衡化、归一化预处理,利用图像阈值分割识别变电缺陷,采用GIS技术定位变电缺陷,完成基于GIS融合的智能变电缺陷运检。试验证明应用本文设计方法,变电缺陷漏检比例和定位相对偏差不超过1%,在智能变电缺陷运检方面应用前景广阔。展开更多
变电站屋面中的保温层、防水层、隔离层、隔汽层和找平层是房屋关键的结构和组成部分,这些层出现的不同表面缺陷,对变电站的性能、使用寿命和人员安全至关重要。本研究基于最优传输分配改进“你仅看一次”(you only look once,YOLOv5s)...变电站屋面中的保温层、防水层、隔离层、隔汽层和找平层是房屋关键的结构和组成部分,这些层出现的不同表面缺陷,对变电站的性能、使用寿命和人员安全至关重要。本研究基于最优传输分配改进“你仅看一次”(you only look once,YOLOv5s)算法来对这些层的表面缺陷进行目标检测,提出了一种更准确和更高效的解决方案,最优传输分配算法通过优化标签分配,提供了比传统阈值方法更精确的匹配,并平衡了正负样本的学习。实验结果表明,最优传输分配优化后的YOLOv5s算法在房屋缺陷的目标检测中能够更全面地考虑图片信息和学习图形特征,减少了定位损失、目标损失和分类损失。此外,最优传输分配还能够提升精确率、召回率和平均准确率(mean average precision,MAP),表明模型的预测准确性、完整性和整体性能得到了改善。因此,使用YOLOv5s算法结合最优传输分配优化的方法对变电站屋面缺陷进行目标检测具有重要的实际应用价值。展开更多
In order to explore the exact nature of deformation defects previously observed in nanostructured Al-Mg alloys subjected to severe plastic deformation, a more thorough examination of the radiation effect on the format...In order to explore the exact nature of deformation defects previously observed in nanostructured Al-Mg alloys subjected to severe plastic deformation, a more thorough examination of the radiation effect on the formation of the planar defects in the high pressure torsion （HPT） alloys was conducted using high-resolution transmission electron microscopy （HRTEM）. The results show that high density defects in the HRTEM images disappear completely when these images are exposed under the electron beam for some duration of time. At the same time, lattice defects are never observed within no-defect areas even when the beam-exposure increases to the degree that holes appear in the areas. Therefore, it is confirmed that the planar defects observed in the HPT alloys mainly result from the significant plastic deformation and are not due to the radiation effect during HRTEM observation.展开更多
文摘变电缺陷运检是智慧变电站的重要组成部分,其可以智能化识别定位变电缺陷,为变电维修提供依据。使用现行方法运检效果不佳,不仅缺陷漏检比例比较高,而且缺陷定位相对偏差比较大,因此本文提出基于地理信息系统(Geographic Information System,GIS)融合的智能变电缺陷运检方法。利用无人机搭载工业相机和GIS设备,感知变电图像和地理信息数据,对图像进行均衡化、归一化预处理,利用图像阈值分割识别变电缺陷,采用GIS技术定位变电缺陷,完成基于GIS融合的智能变电缺陷运检。试验证明应用本文设计方法,变电缺陷漏检比例和定位相对偏差不超过1%,在智能变电缺陷运检方面应用前景广阔。
文摘变电站屋面中的保温层、防水层、隔离层、隔汽层和找平层是房屋关键的结构和组成部分,这些层出现的不同表面缺陷,对变电站的性能、使用寿命和人员安全至关重要。本研究基于最优传输分配改进“你仅看一次”(you only look once,YOLOv5s)算法来对这些层的表面缺陷进行目标检测,提出了一种更准确和更高效的解决方案,最优传输分配算法通过优化标签分配,提供了比传统阈值方法更精确的匹配,并平衡了正负样本的学习。实验结果表明,最优传输分配优化后的YOLOv5s算法在房屋缺陷的目标检测中能够更全面地考虑图片信息和学习图形特征,减少了定位损失、目标损失和分类损失。此外,最优传输分配还能够提升精确率、召回率和平均准确率(mean average precision,MAP),表明模型的预测准确性、完整性和整体性能得到了改善。因此,使用YOLOv5s算法结合最优传输分配优化的方法对变电站屋面缺陷进行目标检测具有重要的实际应用价值。
基金Project (50971087) supported by the National Natural Science Foundation of ChinaProject (BK2012715) supported by the Basic Research Program (Natural Science Foundation) of Jiangsu Province, China+1 种基金Project (10371800) supported by the Research Council of Norway under the NEW Light (NEWLIGHT) Metals of the Strategic Area (SA) MaterialsProject (11JDG070) supported by the Senior Talent Research Foundation of Jiangsu University, China
文摘In order to explore the exact nature of deformation defects previously observed in nanostructured Al-Mg alloys subjected to severe plastic deformation, a more thorough examination of the radiation effect on the formation of the planar defects in the high pressure torsion （HPT） alloys was conducted using high-resolution transmission electron microscopy （HRTEM）. The results show that high density defects in the HRTEM images disappear completely when these images are exposed under the electron beam for some duration of time. At the same time, lattice defects are never observed within no-defect areas even when the beam-exposure increases to the degree that holes appear in the areas. Therefore, it is confirmed that the planar defects observed in the HPT alloys mainly result from the significant plastic deformation and are not due to the radiation effect during HRTEM observation.
