Detecting Real Damage in Operating Tie-Rods Under Uncontrolled Environmental and Operational Conditions

Given the current availability of increasingly advanced sensors and processing capabilities, structural health monitoring (SHM) is the most suitable approach to the maintenance of operating structures. The recent trends in literature indicate that unsupervised data-driven damage detection algorithms may allow a transition from human inspections to continuous and automatic monitoring. However, there are still few examples that prove their effectiveness in real scenarios, i.e., in presence of uncontrolled environmental and operational conditions. This paper tries to bridge this gap, by presenting an application where vibration-based unsupervised damage detection is used to spot the existence of an ongoing corrosion process on operating tie-rods. These structural elements are metallic slender beams subject to axial load, used to balance lateral forces of arches and vaults, in both modern and ancient civil structures. Like all slender structures, they undergo significant vibration levels which make the use of modal identification algorithms particularly effective. In a recent study, the authors demonstrated how modal parameters can be used to define a multivariate damage feature which allows a separation between the environmental effects and damage. This paper investigates the potential of using a Gaussian mixture model for detecting damage in tie-rods through unsupervised data clustering. The potential of the proposed approach is demonstrated considering real data, acquired under uncontrolled environmental and operational conditions, and in presence of real damage.