Early damage detection using multivariate data-driven approaches - Application to experimental data from a cable-stayed bridge

• Tipo: Artigo em Livro de Atas de Conferência Internacional
• Data: 2017

Autores

Sousa Tomé, E (Autor) FEUP; Mário Jorge Pimentel (Autor) FEUP; Figueiras, J (Autor) Outra

Ata de Conferência Internacional

Título: SHMII 2017 - 8th International Conference on Structural Health Monitoring of Intelligent Infrastructure, Proceedings

Páginas: 636-645

8th International Conference on Structural Health Monitoring of Intelligent Infrastructure, SHMII 2017

5 December 2017 through 8 December 2017

Indexação

Scopus - 0 Citações

Outras Informações

ID Authenticus: P-00P-226

Abstract (EN): The implementation of automatic and real-time data processing algorithms in order to reduce the big amount of data down to a human and useful scale is often pointed out as a key step for increasing the value of Structural Health Monitoring (SHM). One of the reasons usually pointed out for the low usability of the structural monitoring data is the fact that damage events are usually masked by the environmental and/or operational effects. Indeed, the robustness and accuracy of the damage (or novelty) detection methods depend on how successfully the changes in the structural response due to damage can be discerned from the normal environmental and operational effects. The process of removing the environmental and/or operational effects from the structural response is usually termed as data normalisation. In this context, the present work describes the adopted methodologies for data normalisation and novelty detection implemented in the Corgo Bridge, a cable-stayed bridge located in northern Portugal recently opened to traffic and wherein a long-term monitoring system has been installed. Data normalisation is accomplished by means of application, alone and combined, of two well-established multivariate statistical tools: multiple linear regression analysis and principal component analysis. The Hotelling T2 control chart is used to track the existence of abnormal values. The performance of the chosen data normalisation methods are evaluated and compared. The first year of data is used to establish the multivariate models and the remaining data is used to validate the fitted models and the ability for novelty/damage detection. Since the bridge is new and sound, damage scenarios are numerically simulated.

Idioma: Inglês

Tipo (Avaliação Docente): Científica