Damage identification of truss bridges based on feature transferable digital twins

Abstract

In this study, a novel damage identification method based on feature transferable digital twins is proposed to improve the damage identification accuracy of truss bridges. Specifically, a finite element model is employed as a benchmark digital model to obtain simulated data for multiple types of truss damage scenarios. The simulated data from the model and the measured signals from the truss bridge are fed into the deep residual sub-domain adaptation network for feature alignment to identify truss bridge damage. The effectiveness of the proposed method is verified by a truss experiment, where the classification accuracies of the training and validation datasets reach more than 80 % and 77 %, respectively. The classification accuracy of the method is maximally improved by 54.51 % compared to conventional identification approaches. The results show that the method has the potential to extract and transfer knowledge of damage patterns from finite element models to identify truss bridge damage.