Structural Modal Calibration of Historical Masonry Arch Bridge by Using a Novel Deep Neural Network Approach

Abstract

The transportation system should be sustained and given serves to improve the well-being of society and continue the improvement of civilization. Historical masonry bridges constitute a critical and sensitive part of the transportation system. As a result of being built a hundred years ago, the bridges have been exposed to many severe deterioration processes and destructive environmental and manmade damage. Therefore, the existing performance of these bridges should be determined realistically in a proper way. This study expresses a novel approach to creating a realistic finite element model of the existing masonry arc bridge. The initial finite element model of the bridge was created according to the architectural drawing of the bridge. Then, the density and elastic modulus of the bridge structural components were investigated statistically and the upper and lower limits were determined. The central composite design approach was used to generate an analytical model cloud, and experimental studies are conducted to determine the real mode shape and frequency of the bridge. Finally, a novel deep neural network approach including deep neural network and principal component analysis-based approaches is proposed to determine the realistic finite element model of the bridge using the results of the analytical models and experimental study. With the proposed methods, the difference between the natural frequency values obtained after the finite element model calibration process and those obtained from experimental measurements was obtained as 1.52% and 0.69% on average in the 6 evaluated mode shapes.