Research On Application Of Damage Identification Method Based On Bridge Monitoring Data

The bridge health monitoring system is an indispensable tool for long-span bridges,and damage identification is an important goal for the development of bridge health monitoring systems.However,with the accumulation of massive data sets,the application of traditional damage identification theories has fallen into a bottleneck in various health monitoring systems,and a large amount of monitoring data has not been thoroughly explored.In order to explore the large amount of hidden information contained in the bridge health monitoring system and improve the shortcomings of traditional structural damage identification methods,this paper proposes a data-driven damage identification method based on monitoring data.Based on Matlab programming,the preprocessing program of bridge monitoring data is developed,and the original monitoring data is effectively processed by credibility evaluation,abnormal data elimination,missing data interpolation and smoothing filter module.Convolution neural network and stack auto-encoding neural network are used to identify the visual image and time series data of monitoring data respectively.The main research work and conclusions are as follows:(1)Taking Mingzhou Bridge as an example,the bridge monitoring data preprocessing program was developed based on Matlab programming,graphical user interface technology and related preprocessing theoretical methods.The core modules include credibility evaluation,abnormal data removal,missing data interpolation and smoothing filter.Grey correlation method is used to evaluate credibility,two-layer elimination method combining probability method and data jump method is used to eliminate abnormal data,Lagrange interpolation and curve fitting interpolation are used to interpolate missing data,and the classic moving average method,five-point quadratic smoothing method and five-point cubic smoothing method are used to filter.(2)After the preprocessing of the monitoring data,Midas was used to establish a finite element model of the Mingzhou Bridge.The vertical acceleration response signal,which is sensitive to structural damage,can reflect structural damage characteristics and is easy to obtain,is selected as the damage index.Then,a damage sample library was constructed,and the damage samples were extracted from the monitoring system and the finite element model.Convolution neural network,stack auto-encoding neural network and corresponding shallow neural network are established by Matlab,which are used for damage identification and classification of monitoring data.The results show that the damage location identification effect based on the deep learning method is significantly ahead of the shallow neural network.The damage identification method based on deep learning has excellent damage pattern memory ability,whether through image identification or data sequence identification,both show excellent performance.The research in this paper can broaden the thinking for dealing with the data redundancy of the bridge health monitoring systems and the research of data-driven damage identification.At the same time,it can also provide reference for the development of generalpurpose bridge monitoring data processing programs.