Research On Structural Health Monitoring Data Cleaning And Early Warning Based On Neural Network

With the development of social economy and information technology,building structures are monitored in real-time by installing various types of sensing devices to monitor their health conditions.Nowadays,the main purpose of structural health monitoring systems is to provide early warning of abnormal structural safety conditions,but many abnormal and distorted data are collected due to power failure,monitoring equipment failure,external environment change,and other factors,which will affect the accuracy of data analysis and early warning results.Therefore,this paper first adopts the corresponding data cleaning method according to the abnormal type of monitoring data,then constructs a structural health monitoring early warning model,and finally designs and implements a structural health monitoring early warning system.The main work of this paper is as follows:(1)By analyzing the monitoring data collected from Nanmao Bridge in Baoting County,Hainan Province,this paper can classify the manifestation of abnormal data into noise,missing,and jumping points.For the cleaning of noisy data,the proposed wavelet transform denoising method and moving average filtering method are stable when dealing with smooth noisy data,but the wavelet transform denoising method performs better when dealing with complex noisy data.For the cleaning of missing data,four processing methods are proposed.Using the mean interpolation method and the direct rejection method has the possibility of losing useful feature information of the data,so they need to be used with caution.The Lagrange interpolation method and ARIMA method are experimentally verified to be more effective in completing the missing data.For the cleaning of jump point data,the Pauta criterion can be used to screen jump points by dividing confidence intervals if the sample data conform to the normal distribution,otherwise,the box plot method can be used to detect jump points.Finally,the NULL value or the average of adjacent data can be set to deal with jump points.(2)By combining the neural network prediction model and the Pauta criterion,an early warning model for structural health monitoring is established.For the problem of the poor prediction effect of a single neural network model,this paper combines CNNLSTM neural network with a self-attention mechanism to establish the prediction model.The self-attention mechanism optimizes the prediction effect of the neural network model by selecting the most critical feature information of the current task.On the basis of this prediction model,the early warning model for structural health monitoring was further constructed by combining the Pauta criterion.The early warning results of bridge displacement data were analyzed according to the evaluation indexes,and the accuracy and completeness rates of the model’s early warning results were 91.75%and 89.00%,respectively.(3)Based on the proposed data cleaning method and early warning model,a structural health monitoring and early warning system was designed and implemented.The monitoring and warning system is designed with five functional modules: user management,data management,data display,data prediction and security warning.In this paper,the various functions of the system are demonstrated and introduced in detail,and the corresponding functional tests are conducted.