Research On Sensor Placement Optimization For Bridge Health Monitoring Based On Deep Learning

Bridge health monitoring must be carried out through a variety of sensors to obtain the required structural information,so the layout of the sensor is one of the key issues,but it should also take into account the economy rather than a large number of layout.In recent years,although some progress has been made in the research of sensor optimal placement,there are still many problems such as complex optimization algorithm and poor optimization results.Based on the design and implementation of the Luding Dadu River Bridge Health Monitoring System and the swarm intelligence algorithm,this paper presents an optimization layout method for the bridge acceleration sensor based on the depth feedforward network.The main contents of the work are as follows:(1)First of all,two aspects of the research on sensor placement and deep learning are reviewed.At the same time,the methods and advantages and disadvantages of the optimal placement of sensors are summarized,and the theory and application of depth learning are also studied.The idea and method of realizing the optimal placement of accelerometers by deep learning are put forward.(2)The direct relationship between the sensor layout scheme and the evaluation criterion(MAC matrix)is established by using the depth feedforward network,and the optimal value and the corresponding optimal layout scheme are obtained by the swarm intelligence optimization algorithm.This paper introduces the whole idea and basic principle of the optimization layout method of acceleration sensor based on deep learning.Then,it introduces the concrete realization process from three aspects: the structural pre processing method of the data,the model establishment of depth learning,the training method and the optimization method of the layout scheme,which are combined with the problem of sensor arrangement of the bridge.(3)Taking the Luding Dadu River Bridge as an example,the finite element model is established by ANSYS software,and the vibration data are obtained.The fixed number of 15 acceleration sensors is arranged on the 54 finite element nodes of the main beam.The optimal parameter range and network structure model suitable for the network are determined through experiments,and the optimization of the network is tested and analyzed.The results show that the average error of the MAC valuepredicted by the layout scheme established by the depth feedforward network and the non diagonal element of the MAC matrix is 0.0055142,and the prediction error of the98.5% arrangement scheme is less than 0.05,and the prediction error of the 88.7%layout scheme is within 0.01.It achieves good training results,and proves that the method is feasible.(4)In the previously trained network,swarm intelligence optimization algorithm is proposed to search the optimal MAC value and corresponding layout scheme of the network.Through genetic algorithm optimization,the optimal MAC value is 0.10815,and the corresponding optimal layout scheme is also obtained.The actual MAC value is 0.11187,the actual error size is 0.00372 and the average error value is less than0.0055142.It is proved that the optimal layout scheme has less error and has better orthogonality between each order.At the same time,the reduction degree is higher than that of the actual vibration mode,which proves that the method can be closely combined with the actual analysis of the health status of the bridge.