Calibration of a reinforced concrete bridge using deep learning

Abstract

Improvement in the assessment of civil structures is an important issues, because the portfolio of bridges in many developed countries keeps aging. This study presents a specific application of neural network for the calibration and the damage assessment of bridges. The focus is on the implementation of the latest state of the art in deep learning to determine and to compute the parameters that fit the finite element model of a bridge to the actual standing structure. Back propagation algorithm with multiple hidden layers is used to train the neural network: The method is first developed in matlab and then deployed in CUDA C. One real bridges are presented, where the method was applied, and one data set of acceleration time histories are analyzed. The obtained autoregressive coefficients are the input of the neural network and the sub element sub matrix scaling factors are the output. Eigenvalue analysis is used to: compare the output of the neural network with the real bridge, and train the algorithm.