Moving Load Identification Of Simply Supported Irder Bridge Based On Dynamic Response

Moving load identification of the bridge is an important part of bridge health monitoring. To obtain accurate and reliable data of the loading can check the load used in the design of bridge. The analysis of load spectrum also can provide basis which is more close to actual for structure fatigue analysis.However, moving load identification of the bridge is immature at present, identification of moving load through vehicle-bridge system model is a deconvolution problem, the mathematical inversion process is mostly illconditioned, so this method is very sensitive to noise.BP neural network is applied to the identification of moving loads on bridges in this paper. Numerical simulation of moving load identification on a simply supported bridge of30m is conducted. The sensitivity of bridge deflection and strain to moving load is analyzed.The influence of different combination of transition function and algorithm on identifing results is discussed. The identifing results in different load case and influence of noise is researched. The applicability of this method is verified by the experimental.The results show that it is feasible to identify moving load of bridge with artificial neural network. Strain response of bridge is more sensitive to moving load than the deflection response. The influence of different combination of transition function on identifing results is less.The maximum mean square error of network is3.7288, the minimum is2.8518,all the correlation coefficient is greater than0.97.However, the influence of training method on identifing results is bigger, the mean square error of network is vary between2.4991and1677.6382, the correlation coefficient is vary between0.1354and0.97717. The recognition accuracy of the load location by the network is better, and the state of the load up and down the bridge and it’s position on the bridges are successfully identified.The maximum error is0.54m. The identification accuracy of the network for wheelbase changes great, and the general rule is the wheelbase bigger and the speed slower, the effect of identification the better,the correct recognition rate of network increased by26.43%when speed is reduced from25m/s to5m/s. When the load is identified, the error of the identification of the vehicle is larger when it’s up and down the bridge than it’s entirely on the bridge, while different wheelbase and speed have a great impact on the identification of the load, the speed greater and the wheelbase greater, the recognition accuracy of the networks the worse, conversely the better; wheelbase has little impact on the accuracy of the speed recognition, and accuracy of the speed recognition is related to the magnitude of the velocity itself, and the greater the speed, the lower the accuracy of recognition. This method has good noise immunity, and in the case of the noise level of20%, the correct recognition rate of network is still greater than60%.The experimental results demonstrate that the relative error of first to forth frequency is respectively5.3%、11.6%、13.5%、15.7，the damping of model beam is light, the first modal damping is0.618%；The maximum recognition error of location is0.464m；The relative error of speed recognition is less than5%. The recognized time history curve of dynamic load fluctuate around static load curve.