基于深度学习和监测数据的桥梁损伤识别方法

为了挖掘桥梁健康监测数据蕴含的大量隐藏信息,以及改进传统结构损伤识别方法的不足之处,提出了基于桥梁监测数据的损伤识别方法。从有限元模拟数据和实际监测数据中分别提取加速度响应,并对原始数据进行了预处理。通过卷积神经网络和栈式自编码网络分别对明州大桥监测数据的可视化图像和时间序列进行识别,同时与浅层神经网络方法的识别正确率对比。结果表明:基于深度学习和监测数据的损伤识别方法不论是通过图像识别还是通过时间序列识别,都表现出优秀的性能:识别正确率达85%以上。与浅层神经网络相比,深度神经网络的损伤工况分类能力更强,识别正确率提高20%以上。

Damage Identification Method of Bridge Based on Deep Learning and Monitoring Data

The study is aimed at revealing the effect of expressway overland runoff particle on the filtration of runoff heavy metals. The gravel - soil filtration columns are used to treat the pavement runoff of Maqun section of Nanjing circle highway. The filtration removal characteristics of heavy metal and the retention characteristics of particulate matter in the expressway overland runoff are studied. The relationship between the retention of particulate matter with different particle sizes and the filtration removal of heavy metal is specially expounded. At the same time, the influences of season, pH value and environmental factors on the removal of heavy metal by gravel-soil filtration are discussed. The results show that Cu and Zn concentrations present the non-linear decrease along the depth direction of gravel - soil filtration column. The 35-cm thick soil above the device has the good interception performance for the removal of the most Cu and Zn. The effect of particle removal on the removal of heavy metals in runoff is improved with the increase of the infiltration padding thickness. The season, pH value, oxidation reduction potential ORP and dry period time have the different levels of influence on the removal of heavy metals.