Many vibration-based damage detection approaches proposed in the literature for civil structures rely on features related to modal parameters, since these are sensitive to structural properties variations. The influence of environmental and operational variability on modal parameters sets limits to unsupervised learning strategies in real-world applications, especially for long-time series. The chapter shows an example of unsupervised learning damage detection in a realistic environment, over a long-time period. Two damage features are compared: one from operational modal analysis and the other from autoregressive models. To start with a real though simple structure, a series of tie-rods has been considered; these are slender axially tensioned beams, widely used in both historical and modern buildings, to balance lateral forces in arches. Since the axial load is heavily influenced by temperature and eventually by other disturbances, even small changes in the environmental conditions cause dramatic changes in the dynamic tie-rod features. To investigate this problem, a set of nominally identical full-scale structures have been continuously monitored for several months under different environmental and operational conditions. It is shown how the combination of vibration-based damage features and multivariate statistics can be successfully used to detect damage in structures working under real environmental conditions.
Vibration-Based Damage Feature for Long-Term Structural Health Monitoring Under Realistic Environmental and Operational Variability
Lucà Francescantonio.;Manzoni Stefano.;Cigada Alfredo
2022-01-01
Abstract
Many vibration-based damage detection approaches proposed in the literature for civil structures rely on features related to modal parameters, since these are sensitive to structural properties variations. The influence of environmental and operational variability on modal parameters sets limits to unsupervised learning strategies in real-world applications, especially for long-time series. The chapter shows an example of unsupervised learning damage detection in a realistic environment, over a long-time period. Two damage features are compared: one from operational modal analysis and the other from autoregressive models. To start with a real though simple structure, a series of tie-rods has been considered; these are slender axially tensioned beams, widely used in both historical and modern buildings, to balance lateral forces in arches. Since the axial load is heavily influenced by temperature and eventually by other disturbances, even small changes in the environmental conditions cause dramatic changes in the dynamic tie-rod features. To investigate this problem, a set of nominally identical full-scale structures have been continuously monitored for several months under different environmental and operational conditions. It is shown how the combination of vibration-based damage features and multivariate statistics can be successfully used to detect damage in structures working under real environmental conditions.File | Dimensione | Formato | |
---|---|---|---|
ch19_def.pdf
Accesso riservato
:
Post-Print (DRAFT o Author’s Accepted Manuscript-AAM)
Dimensione
2.02 MB
Formato
Adobe PDF
|
2.02 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.