Graph-based digital twin modeling for dynamic simulation of continuous beam bridges subjected to unknown loads

Abstract

Digital twin modelling can significantly contribute to accurate and efficient structural analysis and condition identification, whereas existing methods face challenges in simulating structural dynamic responses under unknown load conditions. To address this issue, the present study proposes a Graph-based Digital Twin Modelling (GDTM) method to simulate the dynamic responses of bridge structures without prior knowledge of the external loads. The method uses heterogeneous adjacency matrices to aggregate adjacent measurement responses. A three-span continuous beam bridge and its scaled experimental model are used to validate the proposed method. The results demonstrate that the GDTM method can simulate structural dynamic responses accurately even with unknown loads, achieving a normalized mean squared error (NMSE) of 0.29 for the real bridge and 0.48 for the scaled experimental model, representing an accuracy improvement of over 76 % compared to FEM and 59 % compared to other graph-based methods. Although the model training takes over 15 hours, the simulation takes less than 2 seconds, which is a 28-fold improvement in simulation efficiency compared to FEM. The proposed GDTM method provides a promising solution for digital twin modelling, which may find broad applications in structural operation and maintenance.