A neural network based digital twin model for the structural health monitoring of reinforced concrete bridges

Abstract

Developments in Structural Health Monitoring (SHM) research over the past few decades have demonstrated potential in optimising maintenance solutions for degrading infrastructure. The scale of structural deterioration worldwide and the inadequacy of current non-destructive evaluation techniques necessitate the adoption of accessible, quantitative, continuous SHM technology into mainstream asset management practice. This paper seeks to address this significant demand by proposing a robust, end-to-end, fibre-optic sensor (FOS) monitoring prototype which utilises deep neural networks to convert FOS strain output into an interactive digital twin (DT) visualisation. Finite-element validation demonstrated that the prototype was capable of capturing reliable structural analytics, recording an average error of less than 2kNm and an absolute error of less than 0.15 mm for bending moment and deflection respectively. Furthermore, the predictive mean absolute error of the integrated artificial neural network was less than 1με during testing, demonstrating the accuracy of the digital twin when generating baseline strain data for structural analysis.