Fatigue damage progression is a major concern in the aeronautical and aerospace domains, affecting structure's durability, reliability, availability and safety as well. It can be considered one of the main obstacles to the application of condition-based maintenance or predictive maintenance approaches. For this reason, the scientific community is performing extensive research on structural health monitoring (SHM) systems for the real-time diagnosis of aerospace structures. The diagnostic process aims to identify and localize structural damages, distinguish among several damage types and assess their dangerousness. An SHM system capable of characterizing the state of the structure in real-time is highly desirable so. Despite research on SHM methods requires a considerable amount of further work, another opportunity arises from the availability of real-time SHM data: the real-time prediction of the structure's remaining useful life thus carrying out the prognosis of the system. Research on real-time damage prognosis of aerospace structures has been limited so far, given the complexity of the problem involving (i) an effective diagnostic system, and (ii) interdisciplinary knowledge about structural and damage mechanics, uncertainty representation, uncertainty quantification and theory of stochastic processes. They are the basic tools for a real-time probabilistic framework able to propagate the uncertainty in the future and to evaluate the time-to-failure of the system using effective, quantitative representations like probability density functions. To that end, the thesis investigates the real-time stochastic modeling of fatigue damage progression using a promising sequential Monte Carlo method commonly referred to as particle filter or particle filtering algorithm, suitable for real-time applications. It is a model-based numerical approach to solve the Bayesian prediction-updating problem when other filtering techniques (i.e., Kalman filters) fail because of high-nonlinearity and non-Gaussian probability density functions. First, the development of the algorithm requires a damage propagation model to monitor and to predict the future damage extent caused by fatigue loads. The next step is the embedding of the model into a stochastic framework able to account for the uncertainty of the measurement system and the intrinsic uncertainty affecting the propagation phenomenon. The thesis deals with both fatigue crack growth (FCG) in metallic alloys and progression of multiple damage-modes in composite materials. Literature concerning real-time fatigue damage prognosis and particle filtering is reviewed, and the thesis proposes an enhancement of previous, existing works by providing: (i) the optimal tuning of the stochastic equations underneath the algorithm, which is missing in literature, (ii) a multi-dimensional stochastic framework to predict the remaining useful life of fiber-reinforced laminates affected by multiple damage mechanisms, (iii) a quantitative assessment of the algorithm performance using several experimental FCG data, (iv) the preliminary development of a particle filtering algorithm for random loading-FCG, and (v) a real-time damage detection and monitoring algorithm using a unified, particle filtering-based framework. The advancements and novelties of this work with respect to the state-of-the-art are discussed in the literature review and throughout the thesis, in line with the specific scenario or case study. The efficiency and effectiveness of the algorithm for real-time prognosis is discussed introducing performance metrics from the prognostics health management literature. A critical discussion of all the outcomes, potentiality, advantages and drawbacks of the proposed prognostic framework is provided in the conclusive section. Eventually, future research directions on the topic are discussed, highlighting the author's thought on the future applicability of such algorithms on real aircrafts.
La crescita di danneggiamenti per fatica è una delle problematiche principali nei settori aeronautico e aerospaziale, influenzando la durevolezza, affidabilità, disponibilità e sicurezza di fusoliere ed altri componenti strutturali. Può essere considerata uno degli ostacoli principali all’applicazione di metodi avanzati di manutenzione, come la manutenzione basata sulla condizione (condition-based maintenance), o manutenzione predittiva (predictive maintenance). Per questa ragione, la comunità scientifica ha effettuato e sta tuttora effettuando ricerche estensive nel campo del monitoraggio strutturale o structural health monitoring (SHM) per effettuare, in tempo reale, la diagnosi delle strutture aerospaziali. Il processo di diagnosi ha lo scopo ti identificare e localizzare danni strutturali, distinguendo tra diversi tipi di danneggiamento e valutando la loro pericolosità. Tale sistema di monitoraggio in grado di caratterizzare lo stato della struttura in tempo reale sarebbe, chiaramente, molto vantaggioso per l’industria aeronautica. Nonostante la ricerca nel campo dell’SHM richieda ulteriori, considerevoli studi, la disponibilità di informazioni sullo stato della struttura in tempo reale apre le porte ad una ulteriore possibilità: effettuare la previsione della vita utile residua in tempo reale, effettuando così non solo la diagnosi, ma anche la prognosi della struttura. Le ricerche nel campo della prognosi in tempo reale di danni a fatica sono rimaste limitate fino ad oggi, data la complessità del problema, intrinsecamente multidisciplinare, che coinvolge: (i) un sistema di diagnosi efficace, (ii) conoscenze trasversali di meccanica delle strutture, meccanica della frattura, rappresentazione e quantificazione di incertezze e teoria dei processi aleatori. Questi sono solo gli strumenti di base per un contesto probabilistico in grado di propagare le incertezze nel futuro e stimare il tempo di cedimento del sistema usando rappresentazioni quantitative, come le distribuzioni di densità di probabilità. La tesi investiga la rappresentazione probabilistica della crescita di danni a fatica usando un promettente metodo della famiglia degli algoritmi Monte Carlo sequenziali chiamato particle filtering, particolarmente utile per applicazioni in tempo reale. L’algoritmo si basa su un metodo di filtraggio analitico-numerico per risolvere il problema Bayesiano di previsione e aggiornamento della probabilità in presenza di sistemi altamente non lineari e distribuzioni di probabilitià non Gaussiane, che non è risolvibile con altre tecniche analitiche come il filtro di Kalman. L’algoritmo richiede, innanzitutto, un modello di propagazione del danno per monitorare e prevedere la sua crescita causata da carichi di fatica. Il passo successivo è l’introduzione del modello di propagazione in una struttura probabilistica capace di considerare differenti fonti di incertezza, come il sistema di misura che viene usato per stimare la dimensione del danno e l’incertezza intrinseca che influenza il fenomeno della propagazione stessa. La tesi affronta sia il problema di propagazione di cricche in leghe metalliche (fatigue crack growth, FCG) che la propagazione di molteplici, simultanei danni in materiali compositi. La letteratura riguardante il monitoraggio dei danni usando tale metodo è analizzata nel dettaglio, e la tesi propone ulteriori miglioramenti che riguardano: (i) una discussione critica sull’equazione stocastica alla base dell’algoritmo, che tutt’ora manca in letteratura, (ii) una struttura analitico-numerica multi-dimensionale per la previsione della vita utile di materiali compositi affetti da molteplici danni che interagiscono fra loro, (iii) una valutazione quantitativa delle performance dell’algoritmo utilizzando different test su strutture metalliche reali, (iv) uno sviluppo preliminare dell’algoritmo per il monitoraggio di danni soggetti a fatica ad ampiezza variable ed aleatoria, e (v) una estensione del metodo per l’identificazione di danni ed il loro monitoraggio utilizzando un’unica methodologia per diagnosi e prognosi. Le novità rispetto allo stato dell’arte attuale sono discusse nell’analisi della bibliografia esistente e durante tutta la tesi, in accordo con lo scenario analizzato o i relativi casi esemplificativi. L’efficacia e l’efficienza dell’algoritmo per la prognosi in tempo reale è analizzata utilizzando metriche attualmente esistenti ed utilizzate nel campo del prognostic and health management. La sezione conclusiva fornisce una discussione critica dei risultati, potenzialità, vantaggi e svantaggi della methodologia proposta. Successivamente, le future direzioni della ricerca in questo campo sono discusse, sottolineando il pensiero dell’autore sulla possibile, futura applicazione di questi algoritmi su velivoli reali.
Probabilistic modeling of airframe damage propagation for real-time prognosis
CORBETTA, MATTEO
Abstract
Fatigue damage progression is a major concern in the aeronautical and aerospace domains, affecting structure's durability, reliability, availability and safety as well. It can be considered one of the main obstacles to the application of condition-based maintenance or predictive maintenance approaches. For this reason, the scientific community is performing extensive research on structural health monitoring (SHM) systems for the real-time diagnosis of aerospace structures. The diagnostic process aims to identify and localize structural damages, distinguish among several damage types and assess their dangerousness. An SHM system capable of characterizing the state of the structure in real-time is highly desirable so. Despite research on SHM methods requires a considerable amount of further work, another opportunity arises from the availability of real-time SHM data: the real-time prediction of the structure's remaining useful life thus carrying out the prognosis of the system. Research on real-time damage prognosis of aerospace structures has been limited so far, given the complexity of the problem involving (i) an effective diagnostic system, and (ii) interdisciplinary knowledge about structural and damage mechanics, uncertainty representation, uncertainty quantification and theory of stochastic processes. They are the basic tools for a real-time probabilistic framework able to propagate the uncertainty in the future and to evaluate the time-to-failure of the system using effective, quantitative representations like probability density functions. To that end, the thesis investigates the real-time stochastic modeling of fatigue damage progression using a promising sequential Monte Carlo method commonly referred to as particle filter or particle filtering algorithm, suitable for real-time applications. It is a model-based numerical approach to solve the Bayesian prediction-updating problem when other filtering techniques (i.e., Kalman filters) fail because of high-nonlinearity and non-Gaussian probability density functions. First, the development of the algorithm requires a damage propagation model to monitor and to predict the future damage extent caused by fatigue loads. The next step is the embedding of the model into a stochastic framework able to account for the uncertainty of the measurement system and the intrinsic uncertainty affecting the propagation phenomenon. The thesis deals with both fatigue crack growth (FCG) in metallic alloys and progression of multiple damage-modes in composite materials. Literature concerning real-time fatigue damage prognosis and particle filtering is reviewed, and the thesis proposes an enhancement of previous, existing works by providing: (i) the optimal tuning of the stochastic equations underneath the algorithm, which is missing in literature, (ii) a multi-dimensional stochastic framework to predict the remaining useful life of fiber-reinforced laminates affected by multiple damage mechanisms, (iii) a quantitative assessment of the algorithm performance using several experimental FCG data, (iv) the preliminary development of a particle filtering algorithm for random loading-FCG, and (v) a real-time damage detection and monitoring algorithm using a unified, particle filtering-based framework. The advancements and novelties of this work with respect to the state-of-the-art are discussed in the literature review and throughout the thesis, in line with the specific scenario or case study. The efficiency and effectiveness of the algorithm for real-time prognosis is discussed introducing performance metrics from the prognostics health management literature. A critical discussion of all the outcomes, potentiality, advantages and drawbacks of the proposed prognostic framework is provided in the conclusive section. Eventually, future research directions on the topic are discussed, highlighting the author's thought on the future applicability of such algorithms on real aircrafts.File | Dimensione | Formato | |
---|---|---|---|
MatteoCORBETTA_PhD_thesis_XXVIII.pdf
non accessibile
Descrizione: MatteoCorbetta_PhDThesis_XXVIII
Dimensione
118.92 MB
Formato
Adobe PDF
|
118.92 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/10589/122301