摘要
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。
In order to master a large amount of hidden information contained in bridge health monitoring data and improve the shortcomings of traditional structural damage identification methods,a damage identification method based on bridge monitoring data is proposed.The acceleration response is extracted from the finite element simulation data and the actual monitoring data.And the raw data are preprocessed.By using the convolution neural network and stacked auto-encoding neural network,the visual images and time series of monitoring data of Mingzhou Bridge are identified respectively,at the same time,are compared with the identification accuracy of the shallow neural network.The results show that the damage identification methods based on deep learning and monitoring data all have the excellent performances whether through the image identification or through the data sequence identification.The identification accuracy rate is over 85%.Compared with the shallow neural network,the deep neural network has the stronger ability to classify the damage conditions,and the identification accuracy rate is increased by more than 20%.
作者
唐良
边祖光
赵银飞
金婉
TANG Liang;BIAN Zuguang;ZHAO Yinfei;JIN Wan
出处
《城市道桥与防洪》
2022年第1期174-180,M0016,M0017,共9页
Urban Roads Bridges & Flood Control
基金
国家自然科学基金项目(11572286)。
关键词
损伤识别
深度学习
监测数据
卷积神经网络
栈式自动编码器
damage identification
deep learning
monitoring data
convolutional neural network
stacked autocoder
