基于径向基神经网络的桥梁有限元模型修正

基于某预应力混凝土大跨刚构-连续梁桥的ANSYS有限元模型,提出一种基于径向基神经网络的有限元模型修正方法。该方法以不同设计参数条件下有限元模型模态分析频率作为输入向量,以对应的桥面单元、中墩、边墩的弹性模量、密度等设计参数修正值作为输出向量,利用径向基神经网络来逼近两者之间的非线性映射关系。结合该桥梁结构健康监测系统中加速度传感器监测的桥梁结构动力反应的加速度数据,利用神经网络的泛化特性,直接计算出有限元模型设计参数的修正值。研究结果表明:修正后的有限元模型能更真实地反映结构的物理状态,较好地反映该桥梁结构的真实动力特性。

Bridge model updating based on radial basis function neural network

A model updating method based on radial basis function neural network(RBF) is proposed for an ANSYS finite element model of a prestressed large span rigid-continuous concrete bridge.This method utilized the finite element modal analysis under different design parameter conditions as input vector,and the elastic modulus,density of box girders,piers as output vector.The nonlinear relationship between the inputs and outputs is approximated through RBF.With the generalization of neutral network,combined with bridge dynamic response monitored by acceleration sensors from the bridge structural health monitoring system,the corrected value of the finite element model of the design parameters is calculated.The research shows that the updated model can present the true physical conditions and the results of updated model better reflect dynamic characteristics of the bridge structure.