Damage Localization and Quantification as a Bayesian Inverse Problem in Structural Health Monitoring

Abstract

The maintenance of highways and railways requires timely and cost-effective bridge monitoring. However, conventional inspections are performed manually and become increasingly time-consuming as structures age. We suggest an interpretable, output-only structural health monitoring approach based on acceleration data, a physics‑based beam model and a Bayesian model updating framework, where damage detection is cast as an inverse problem of parameter estimation. The linearized Bayesian inference scheme allows to identify deviations in material properties and thus to determine both the localization and quantification of damage as well as uncertainty ranges. By linking the physical asset with its digital counterpart, this method is an important component of a hybrid digital twin.