Model correction of continuous girder bridges based on neural network algorithm and natural frequency testing
In order to solve the problem of deviation between the finite element model results and the actual structure,this paper takes continuous girder bridges as research object and proposes a corrective approach for finite element models based on neural networks and natural frequency testing.Firstly,the fundamental principles governing backpropagation neural networks(BPNN),radial basis function neural networks(RBFNN),and particle swarm optimization-backpropagation neural networks(PSO-BPNN)are introduced.The finite element models for three types of continuous girder bridges featuring distinct cross-sections:hollow slabs,box girders,and T-beams are then established and analyzed.This involves computing the corresponding dynamic characteristics and conducting vibration tests on real bridges.Sensitivity analysis is employed to identify parameters requiring correction.A comparative assessment of the corrective impacts of these three neural networks ensues,validated against empirical data.Notably,the elastic modulus and density exert a more pronounced influence on natural frequencies for bridges compared to Poisson's ratio,prompting their selection for correction.The corrected finite element models exhibit errors reduced to within 7%,accompanied by all corrected modal assurance criterion(MAC)values surpassing 94%.PSO-BPNN demonstrates superior accuracy and reliability when juxtaposed with BPNN and RBFNN,yielding finite element models that more accurately depict the mechanical response of actual bridge structures under realistic conditions.
万方数据
维普
