| The main purpose of bridge health monitoring is to collect various static and dynamic responses of the structure,such as structural deflection,strain,acceleration,crack width and traffic load,etc.,by building a bridge health monitoring system,and to evaluate the structural state based on these response data.These response data contain the state information of the structure,and the establishment of its benchmark prediction model is crucial to the state evaluation of the structure.In order to effectively use the bridge monitoring data and explore the structure’s own state information contained in this paper,this paper innovatively proposes that a neural network prediction model based directly on the monitoring data can be established between different types of response of the bridge structure,so as to achieve a class of response Prediction of another type of response.In this paper,four types of neural network prediction models are established and the prediction performance,noise adaptability and noise reduction methods of the model are studied.The four types of models include: 1)deflection-load position neural network model;2)strain-load position neural network Network model;3)strain-deflection neural network model;4)strain-fracture width neural network model.The research in this paper is completed by a combination of numerical simulation and real bridge verification.The main research contents include:(1)Use the finite element software Abaqus to establish a three-dimensional finite element analysis model of the three-span continuous box girder bridge with and without cracks,and generate the strain,deflection and cracks of the simulated measuring points under static and dynamic loads through numerical analysis Width response data to form various sample data sets required for subsequent neural network models;(2)Establish a deflection-load position prediction model based on BP neural network under static load conditions to realize the use of structural deflection to predict the position of vehicle load.The influence of noise on the prediction performance of the neural network model is studied.Finally,the mean filtering method is introduced into the model to reduce the noise of the sample data;(3)A strain-load position prediction model based on BP neural network is established to realize the use of structural strain response to predict the position of vehicle load.The influence of noise on the prediction performance of the model and the method of reducing the influence of noise are studied;(4)Establish a strain-deflection prediction model based on BP neural network under static and dynamic loading conditions respectively,to realize the prediction of the structure deflection from the structural strain response,and study the effect of noise;(5)Establish the strain-fracture width prediction model based on BP neural network under static and dynamic loading conditions respectively to realize the prediction of structural crack width from structural strain response data,and further study the model’s ability to adapt to noise.(6)The method proposed in this paper is used for real bridge verification,and a strain/temperature-deflection prediction model is established for Wanzhou Bridge,and the prediction accuracy of the model is tested.The research results in this paper show that the above prediction models based on BP neural network can accurately predict the vehicle load position from the strain or deflection of the structure,and the deflection and crack width of the structure from the strain/temperature of the structure,especially when the mean filtering method is introduced After performing noise reduction processing on the sample data,the prediction accuracy of the model is further improved,showing good noise adaptability.The various prediction models based on BP neural network developed in this paper can be used as a benchmark model for bridge health monitoring,which can provide an important basis for the condition assessment of bridge structures and also help to solve the problem of missing monitoring data in health monitoring systems.In addition,because the model is directly based on the bridge health monitoring data and avoids the bridge’s mechanism model,it is expected to play an important role in bridge health monitoring. |
