Wheel Force Identification Based On Multi-sensorin Rapid Bridge Test

As an important traffic structure across rivers or other geographical obstacles,in order to ensure the safety of such a number of bridges,the performance evaluation of bridges is of great significance.Impact vibration is one of the main means of structural health monitoring.Compared with other means,its in-out can be identified is an important advantage,so that the deep parameters of the structure can be identified.At the present stage,in order to avoid the limitation of "stop and excitation",an ideal way is to use the vehicle itself as the excitation directly.Therefore,tire is the only component of vehicle-bridge contact,and vertical wheel force is the input force of impact vibration.Its accurate identification plays an important role in the rapid health monitoring of bridges.Although there are many methods of wheel force measurement,they can not be used in bridge rapid testing because of the high cost of sensors or other reasons.In this paper,an intelligent tire vertical wheel force recognition method based on multi-sensor information fusion is proposed.Compared with the traditional tire model,this method takes into account the tire parameters such as vertical deformation,tire pressure and rotational speed,and converts them into a time series prediction problem to accurately identify the vertical wheel force.After that,an intelligent tire tester is designed to verify the validity of the proposed neural network algorithm model,experiments are designed based on the relationship between single parameter of tire and wheel force or multiparameter coupling of tire and wheel force.The main contents of this paper are as follows:(1)Based on the mechanical model of tire,the applicable models of various tires under dynamic state are discussed.Because of the limitation of traditional models in fitting vertical wheel force under multi-parameter coupling change,this paper presents a detailed analysis based on BP neural network algorithm and LSTM deep learning algorithm.The matching problem between sensor data acquisition and input-output layer of neural network is solved,Including the phenomenon of uneven tire pressure caused by tire rotation,which need to collect tire pressure data by using demy wavelet filter,and the phenomenon of the pre-contact between tire and obstacle,which need to collect laser displacement sensor data by using the concept of "effective height".(2)The intelligent tire tester and related tire tests are introduced.The single-factor and multifactor coupling changes between tire pressure,tire speed,tire vertical deformation and vertical wheel force are experimented.The fitting effects of tire model based on BP neural network algorithm and LSTM depth learning algorithm are compared,and some errors between laboratory test and real engineering are discussed and analyzed.The main conclusions are as follows: BP algorithm neural network has good fitting ability when tire pressure and speed change,but it can not get accurate fitting results when obstacle height changes;LSTM depth learning algorithm neural network,through sequential samples,does not need to transform the vertical deformation,and in various working conditions of tire,fits accurately.Fitting accuracy is maintained at a very high level,although some high-frequency "burrs" will be generated in some segments,but they all fluctuate within an acceptable range.(3)For the rapid test of bridges,the information of road surface roughness cannot be measured in advance,and the shape of obstacles cannot be a simple and identical shape.Therefore,the calibration of "effective height" of obstacles cannot be completed in advance,so the BP neural network algorithm with better simulation results in the laboratory cannot be directly applied to the engineering test;LSTM depth learning algorithm based on sequential samples can directly use monitoring data to train and learn in the input layer of neural network by selecting reasonable samplelength without calibrating road surface roughness.The average error of LSTM depth learning algorithm also meets the accuracy requirement,and is not greatly affected by the change of working conditions,so it has good engineering application prospects.