Structural health monitoring (SHM) is an important topic in civil engineering due to the great significance of infrastructures such as high-rise buildings, bridges, dams, etc. The main objective of SHM is to evaluate the integrity and safety of these structures for early damage detection, damage localization, and damage quantification. Recently, data-driven methods based on statistical pattern recognition present efficient approaches to detect damage by using measured vibration data. These methods are generally based on two main steps, including feature extraction and statistical decision-making. Time series analysis and distance- based approaches are effective tools for these steps. Despite advantages of data-driven damage diagnosis methods, those may give unreliable results of damage diagnosis in terms of false alarm and false detection errors. These errors may be caused by some challenging issues including inappropriate feature extraction resulting from insufficient and inaccurate time series modeling under non-stationary vibration signals caused by ambient excitations, inaccurate statistical decision-making for damage detection due to the effects of environmental and operational variability and high-dimensional features. To deal with these issues, this dissertation proposes novel data-driven methods in the steps of feature extraction and statistical decision- making. The proposed approaches are related to innovative techniques for model order determination, residual-based feature extraction methods based on time series modeling, hybrid methods for feature extraction under ambient excitations and non-stationary vibration signals, several distance-based approaches to statistical decision-making for early damage detection, damage localization, and damage quantification. The effectiveness and accuracy of the proposed methods are validated by numerical and experimental structures, moreover several comparative are reported. Results demonstrate that the methods proposed in this dissertation are effective and robust tools for SHM of civil structures under environmental and operational variability conditions and ambient excitations.
Il monitoraggio strutturale (in inglese Structural Health Monitoring, SHM) è un argomento di ricerca importante per l'ingegneria civile, a causa del grande valore di infrastrutture come grattacieli, ponti, dighe, ecc. L'obiettivo principale del monitoraggio è la valutazione dell'integrità e della sicurezza di queste strutture rilevando, localizzando e quantificando possibili danni. Recentemente, metodi basati soltanto sui dati (che non necessitano cioè di modelli fisici) e modelli statistici sono emersi come approcci efficienti per rilevare i danni, utilizzando informazioni estratte dalle vibrazioni strutturali. Questi metodi si differenziano generalmente in due fasi principali: estrazione delle caratteristiche (feature extraction); processo decisionale su base statistica (statistical decision-making). L'analisi delle serie storiche e gli approcci basati sulla distanza o variazione tra loro sono strumenti efficaci; nonostante i vantaggi di questi metodi, questi possono fornire risultati inaffidabili, in termini di diagnosi dei danni, che danno luogo a falsi allarmi o falsi errori di rilevazione. Questi errori possono essere causati da una estrazione inappropriata delle caratteristiche, derivante da una modellazione insufficiente e imprecisa come serie temporale in caso di vibrazioni non stazionarie dovute a eccitazioni ambientali, oppure da un processo decisionale inaccurato per il rilevamento dei danni, a causa degli effetti della variabilità ambientale e delle condizioni operative o per caratteristiche di grandi dimensioni (Big Data). Per affrontare questi problemi, questa tesi propone nuovi metodi per le fasi di estrazione delle caratteristiche e di decisione statistica. Gli approcci proposti sono basati su tecniche innovative per: la determinazione dell'ordine dei modelli; l’estrazione di caratteristiche basata sui residui della modellizzazione di serie temporali, con metodi ibridi sotto eccitazioni ambientali e segnali non stazionari; il calcolo della distanza o variazione tra lo stato corrente e quello non danneggiato di riferimento, utilizzato per il processo decisionale atto al rilevamento precoce, alla localizzazione e alla quantificazione del danno. L'efficacia e l'accuratezza dei metodi proposti sono verificate tramite esempi numerici e reali, proponendo anche analisi comparative con metodi allo stato dell’arte. I risultati dimostrano che i metodi proposti in questa tesi sono efficaci e robusti per il monitoraggio di strutture civili, in condizioni di variabilità ambientale e operativa e sottoposte ad eccitazioni ambientali.
Vibration-based structural health monitoring by novelty detection and feature extraction techniques
ENTEZAMI, ALIREZA
Abstract
Structural health monitoring (SHM) is an important topic in civil engineering due to the great significance of infrastructures such as high-rise buildings, bridges, dams, etc. The main objective of SHM is to evaluate the integrity and safety of these structures for early damage detection, damage localization, and damage quantification. Recently, data-driven methods based on statistical pattern recognition present efficient approaches to detect damage by using measured vibration data. These methods are generally based on two main steps, including feature extraction and statistical decision-making. Time series analysis and distance- based approaches are effective tools for these steps. Despite advantages of data-driven damage diagnosis methods, those may give unreliable results of damage diagnosis in terms of false alarm and false detection errors. These errors may be caused by some challenging issues including inappropriate feature extraction resulting from insufficient and inaccurate time series modeling under non-stationary vibration signals caused by ambient excitations, inaccurate statistical decision-making for damage detection due to the effects of environmental and operational variability and high-dimensional features. To deal with these issues, this dissertation proposes novel data-driven methods in the steps of feature extraction and statistical decision- making. The proposed approaches are related to innovative techniques for model order determination, residual-based feature extraction methods based on time series modeling, hybrid methods for feature extraction under ambient excitations and non-stationary vibration signals, several distance-based approaches to statistical decision-making for early damage detection, damage localization, and damage quantification. The effectiveness and accuracy of the proposed methods are validated by numerical and experimental structures, moreover several comparative are reported. Results demonstrate that the methods proposed in this dissertation are effective and robust tools for SHM of civil structures under environmental and operational variability conditions and ambient excitations.File | Dimensione | Formato | |
---|---|---|---|
Dissertation_Alireza Entezami_Final.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Dissertation_Alireza Entezami
Dimensione
16.66 MB
Formato
Adobe PDF
|
16.66 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/166650