Research On Fault Diagnosis Of Bridge Health Monitoring Sensor Based On Deep Learning

Various bridge parameters could obtained by the bridge health monitoring real time,and the condition of the bridge structure is judged,prompt repair measures is making to prevent the large-scale structural damage appearance at bridge,so as to ensure that the internal structure of the bridge does not exceed the damage limit.How to obtain accurate bridge parameter data is the basis of the bridge health monitoring system,the accuracy of the data directly determines the judgment of bridge health monitoring system's about the bridge status.The direct means of obtaining monitoring data is by sensors,but it works in a harsh environment and work at high intensity,so they are prone to failure.Based on this,based on deep learning theory proposed a sensor fault diagnosis and fault type identification method.First,first uses the deflection monitoring in a continuous rigid frame bridge as an example to discuss the working principle of the laser displacement displacement sensor for monitoring deflection,and the types of failures that may occur under the influence of the working environment of the bridge and the natural environment of the laser projection displacement sensor.Failure characteristics and the cause of the failure.Then,by comparing several kinds of deep learning models,the depth self-encoding network is selected as the fault diagnosis model,and the L-BFGS optimization algorithm is selected as the model optimization algorithm.Combining the fault data characteristics of bridge sensors,the input and output neurons of the deep self-encoding network are determined,and the number of deep self-encoded network layers and the number of hidden layer nodes are determined by offline simulation experiments.And,the proposed fault diagnosis method was validated by using the measured data of the bridge,and being compared with the traditional BP neural network sensor fault diagnosis method and the support vector machine sensor fault diagnosis method.The comparison results showed that the diagnostic accuracy rate of BP neural network is 74.5%.The diagnostic accuracy rate of the support vector machine is 76.7%,the diagnostic accuracy of the depth self-encoded network is 94.5%,and the accuracy of the sensor fault diagnosis method based on deep learning is improved by 17.8%,which could solve the problem of low sensor fault diagnosis rate very well.Finally,the characteristics of bridge damage are analyzed,and a method of distinguishing bridge damage and sensor failure is constructed based on sensor correlation.