Toward digital twin-enabled real-time structural damage detection via POD-assisted Bayesian estimation

Accurate and timely detection of structural damage is essential for ensuring the safety and integrity of civil infrastructure. While Bayesian filtering has been widely adopted in structural health monitoring (SHM) for estimating system states and parameters, full-scale parameter estimation can often result in biased outcomes and significant computational costs. To overcome these limitations, this study proposes a framework toward digital twin-enabled SHM by integrating proper orthogonal decomposition (POD) with Bayesian filtering. The framework uses displacement and excitation measurements from the system to estimate states in real-time, updated proper orthogonal mode (POM) slopes, and the location and severity of damage. Structural damage localization is achieved by continuously monitoring changes in the slope of the first POM, which is updated in real time using an online singular value decomposition (SVD) algorithm. Simultaneously, the Unscented Kalman Filter (UKF) estimates system states and selectively updates stiffness parameters, focusing only on degrees of freedom identified as damaged by comparing the evolving POM slopes with a baseline reference derived from the healthy structure. This coordinated operation of online SVD and UKF enables efficient, accurate, and robust assessment of structural condition in real time. The proposed framework addresses three critical stages of SHM that is, detection, localization, and quantification. Its performance is validated through comprehensive numerical simulations on an eight-story shear frame under various damage scenarios, including adjacent-story, non-adjacent, and progressive damage cases. Furthermore, validation on a laboratory-scale three-dimensional steel frame confirms its practical applicability. The results demonstrate that the proposed approach can reliably localize and quantify multiple damage events, offering a scalable solution for real-time SHM.