Eigenfrequency-Based Feature for Automatic Detection of Real Damage in Tie-Rods Under Uncontrolled Environmental Conditions

Data-driven approaches to damage detection are very common in the field of structural health monitoring (SHM) due to the possibility to be adopted without the need for a model of the monitored structure. These approaches rely only on the information contained in data acquired by sensors, which must be extracted through the adoption of appropriate damage features, i.e., synthetic indexes highly correlated to the structural state. In an unsupervised learning perspective, i.e., when data referring to damage are not available prior to the monitoring, the damage is detected when the damage feature shows a significant variation with respect to a reference set, containing data referring to the initial healthy condition. A critical aspect for unsupervised learning data-driven approaches is related to the fact that, usually, changes of damage features due to environmental and operational variations (EOVs) can be greater than those caused by damage. Many approaches are proposed in the literature to tackle this problem, most of which are validated on simplified cases, under controlled laboratory conditions and where, often, the effect of the damage is only simulated, resulting in a difficult translation to real applications. In this work, attention is paid to the development of an automatic damage detection strategy for axially loaded beam-like structures, that can be used without the supervision of an expert and that allows for identifying a real state of damage, under the effects of an uncontrolled environment. More in detail, the case study of tie-rods (i.e., tensioned metallic beams used to balance lateral forces of arches and walls of civil structures) is addressed. In previous work, the authors showed that when multiple vibration modes are considered together, patterns of modal parameters associated with damage are different from those due to the effects of EOVs, allowing for defining effective damage features. In this chapter, the strategy is further developed, with a focus on the automatization of the strategy for real applications. This means not only dealing with EOVs, but also developing a successful automatic data cleansing strategy, to automatically detect and discard corrupted results obtained when the operational modal analysis algorithms fail due to unfavorable operating conditions. The validity of the proposed framework is demonstrated on real long-term monitoring data and in presence of real corrosion damage, which is a rare case-study in the field of SHM.