Research On Structural Strain Threshold Setting Of Bridge Health Monitoring System Based On Deep Learning

During the construction and operation of bridges,they are often subjected to loads and environmental erosion,which can affect the safety status of bridge structures.Therefore,it is necessary to monitor the bridge structure,understand its safety status,and ensure the safety of people’s lives and property.Due to its ability to monitor the safety status of bridge structures in real-time,more and more bridges are choosing to install bridge health monitoring systems.This article focuses on the method of setting structural strain thresholds for bridge health monitoring systems.Firstly,the main denoising methods for monitoring data at home and abroad were studied,and a new soft hard compromise function was proposed based on the wavelet threshold denoising method;And the sparrow search algorithm is used to automatically optimize and determine the adjustment factor of the function,making the value of the adjustment factor more flexible and more suitable for the data,thereby improving the overall denoising effect of the method.Then,a joint LSTM-Markov strain prediction and threshold setting model is constructed based on the deep learning neural network short and long term memory neural network(LSTM)and the machine learning model Markov model.The noise reduced data is used to predict the monitoring index threshold and set the threshold.Finally,the method proposed in this article is validated through the Lanjiawan High Pier Steel Trestle Health Monitoring System,based on the collected actual engineering strain monitoring data.The specific work content is as follows:(1)A universal threshold calculation regulation based on wavelet threshold denoising method was selected,which is more in line with the characteristics of wavelet decomposition.The compromise function between soft and hard thresholds was improved to construct a threshold function with wider applicability and better denoising effect.The fitness function of the sparrow search algorithm is constructed by using the signal to noise ratio(SNR),the evaluation index of the denoising effect.The sparrow search algorithm is used to optimize the adjustment factor of the threshold function,which changes the way of relying on experience to obtain the adjustment factor through trial and error,making the value of the adjustment factor more flexible,more consistent with the data characteristics,and further improving the denoising effect.Finally,a simulation signal was constructed using Matlab and denoised to verify the feasibility of the above method.(2)Propose a joint LSTM-Markov strain prediction model to directly predict the strain data transmitted by the bridge health monitoring system,and obtain the predicted strain values of the bridge structure at future times.Based on the relationship between the predicted values of the previous joint strain prediction model and the actual values,the reasonable fluctuation range of the future strain values is calculated based on the concept of confidence intervals,which is the strain monitoring threshold of the bridge health monitoring system in the future.Abandoning the previous strain dynamic threshold setting method of separating and then stacking bridge strains,overcoming the shortcomings of this strain threshold setting method that cannot meet the requirements of engineering practice for timeliness and the complexity of calculation.The use of Fisher ordered clustering method to improve the Markov model solves the problem of interval partition number and interval width not being dynamically adjusted with actual data characteristics in the equidistant partition state of the Markov model.And write the above methods into code to achieve automatic setting of dynamic thresholds for monitoring indicators in the bridge health monitoring system.(3)Based on the practical application of the Lanjiawan High Pier Steel Trestle Health Monitoring System project,the feasibility of the methods described in Chapter 2 and Chapter3 of this article was verified,and the prediction effects of using the combined LSTM-Markov strain prediction model and using LSTM prediction alone were compared.Firstly,a finite element model of the bridge was established using Midas Civil software,and the strain threshold set according to the specifications and model method was obtained.Secondly,denoise the bridge strain monitoring data.And using the joint model to obtain the predicted strain value of the bridge,based on the idea of confidence intervals,the strain prediction value of the Lanjiawan high pier steel trestle bridge is scaled down to obtain the strain threshold of the bridge health monitoring system.Finally,by comparing the strain thresholds set according to the specifications and model methods,it was found that the threshold set in this method is more in line with the law of strain changes,and the threshold setting is more reasonable.And combine the two to establish a three-level warning system for strain threshold setting.