基于深度学习和IHPO的桥梁结构模型修正方法

Bridge structure model updating method based on deep learning and IHPO

针对桥梁结构模型修正中实测振型不完备的问题,基于简支梁动力试验,利用卷积神经网络强化实测非完备振型,联合强化后的完备振型和实测频率构建目标函数,采用改进的猎食者优化算法修正结构有限元模型,并与用实测非完备振型和频率构建目标函数进行模型修正的方法进行对比。结果表明:基于深度学习和改进猎食者优化算法修正的简支梁结构有限元模型的前4阶频率与实测值误差均小于0.5%,振型的模态置信准则值均大于0.99,且模型修正的平均消耗时间比使用实测非完备振型进行模型修正时少45%;所提出的方法具有更高的精度和效率,且传感器数量仅为4个时12单元简支梁也能获得精确的模型修正结果。

Aiming at the problem of incomplete measured mode shapes in bridge structure model updating,based on the dynamic test of a simply supported beam,the convolutional neural network was used to enhance the measured incomplete mode shapes,and the objective function was constructed by combining the enhanced complete mode shapes and the measured frequencies.The improved hunter prey optimization algorithm was applied to update the structural finite element model,which was compared with the model updating method by using the measured incomplete mode shapes and frequencies to construct the objective function.The results indicate that the errors between the first four frequencies and the measured values of the finite element model of the simply supported beam structure updated based on deep learning and improved hunter prey optimization algorithm are less than 0.5%,and the model assurance criterion values of mode shapes values are greater than 0.99.The average consumption time of the model updating is 45%less than that of the model updating using incomplete mode shapes.The proposed method has higher accuracy and efficiency,and even when the number of sensors is only 4,the 12-element simply supported beam can also obtain accurate model updating results.