Study Of The Multi-vehicle Bridge Weigh-in-motion Algorithm Based On The Convolutional Neural Network

With the increasing demand for highway freight,both of the tonnage of road freight trucks and the total mileage of roads increase,causing the common and serious overloading problems.Vehicle overloading could not only cause irreversible damage to the bridge structure,but also accelerate the bridge degradation,and even directly lead to the bridge collapse,which seriously affects the safety and sense of safety of the public.Vehicular overloading has become the leading human factor that causes bridge failure,and has been a major safety hazard to the safety of highway bridges in our country.In this regard,identifying the load information of passing vehicles timely and accurately and controlling the overloaded truckss reasonably and effectively are significant to the safety of highway bridges.Since the bridge-weigh-in-motion(BWIM)technology could identify the weight information of passing vehicles quickly while not interrupting the traffic,it has been widely used in the engineering and has become an important means of identifying and weighing overloaded vehicles.However,the Moses algorithm,which is applied widely in the BWIM technology mainly concentrated on the fact that only one vehicle pass the bridge,and failed to consider the effect of the vehicle lateral loading position on the identification results,which therefore lead to the bottleneck of relative low recognition accuracy under the multi-vehicle loading conditions.In order to consider the multi-vehicle loading conditions and therefore increase recognition accuracy,the improvement for the BWIM technology mainly focused on calibrating the accurate bridge influence surface.However,the challenges still existed,i.e.,the difficulty and high requirements to obtain the accurate influence surface.In this regard,the deep learning method was adopted in this study to develop the convolutional-neural-network-based multi-vehicle bridge-weigh-in-motion algorithm to timely and accurately measure the vehicular weight information,i.e,the axle load and total weight under the condition of multi-vehicle loading in the transverse direction of the bridge.The main contents were as follows:(1)The theory and corresponding principle were comprehensively reviewed and analyzed regarding the classical bridge-weigh-in-motion method based on the Moses algorithm and its derivative algorithms,the BWIM method that utilized the influence area information to calculate the total vehicle weight directly,and the innovational BWIM method based on the neural network.In addition the corresponding advantages and disadvantages for such methods were also discussed.(2)A new BWIM algorithm based on convolution neural network for axle weight identification of multiple vehicles was preposed.The vehicle-bridge coupling system was used to collect the bridge strain signals under different loading conditions that vehicle traveled with different speeds,numbers,loading positions and weights.And then the convolutional neural network was utilized to learn the variation pattern of the strain signals collected from the bottom of the beam under the conditions of vehicle loading with different model parameters.Finally,the structure of the convolutional neural network was optimized.In addition,the axle load recognition accuracies for the single vehicle and multiple vehicles,which traveled with different speeds and and crossed the bridge following different lateral positions,were compared and analyzed for the developed BWIM algorithm(3)Based on the fully use of the correlation,as contained in the trained convolutional neural network model in the previous part,between the bridge responses and the vehicle axle load.A transfer learning method was ingrated with the developed BWIM algorithm to reduce the train data when applied to other bridges.Moreover,the stability and identification accuracy of the updated BWIM technology based on the transfer learning method and the developed BWIM technology in previous part were compared and analyzed considering different loading conditions,i.e.,different transverse loading positions,different vehicle types and different traveling speeds.