The recent advances in materials and manufacturing technology have revolutionized the field of structural engineering, aiming for increasingly efficient and high-performance structures. The workhorse of this revolution is the category of fiber reinforced plastics (FRP), exhibiting superior capabilities in terms of specific stiffness and strength and making them particularly appealing in several structural applications where weight savings are crucial. However, FRP are notably vulnerable to flaws developing from the manufacturing process and service, and unlike metals, are featured by complex and multivariate failure and degradation modes, leading to uncertainty in the assessment of current and future material properties. These shortcomings have prevented the fielding of systematic predictive maintenance (PM) strategies for complex, large scale FRP structures, which, at best, rely on resource-in-tensive periodic inspections by non-destructive evaluation (NDE) techniques. With increasing use of FRP in several industries such as aviation, effective reliability analysis of composite structures has become imperative in recent years. This desideratum may be fulfilled by developing ad-hoc damage prognosis (DP) methods. DP methods attempt to predict the remaining time beyond which the structure is expected not to perform within desired specifications, i.e., the remaining useful life (RUL), by inferring the current state of the system (i.e., structural health monitoring), estimating its future operational environments, and then predicting its RUL. In practice, DP requires the integration of diverse fields of technology and a variety of predictive modelling approaches, including the ability to quantify the uncertainty associated with RUL predictions, to enable informed, risk-based maintenance decisions. The increased ability to sense, acquire and process data has accelerated research and development breakthroughs in the field data-driven (DD) approaches, envisaging a data-centric future for many scientific disciplines, including DP. In the context of DP of composite structures DD approaches have been strongly encouraged, as a universal structure underlying damage progression models is missing, and the field suffers from ambiguous and inconsistent interpretations. It is however questionable whether the success of DD approaches in domains where data are abundant and physics-based models do not exist should trace the route of DP, being a field where data are typically scarce and scientific knowledge defining how processes evolve is, to some extent, available. The hesitation is even more legitimated by considering that DP requires a predictive effort, and purely DD approaches are unlikely to be predictive, having intrinsic limitations on the cognitive tasks that they can perform. The goal of this research is to develop a framework for the DP of FRP structures that merges the partial knowledge of condition monitoring (CM) data from an integrated sensors network with the limited predictive abilities of damage progression models, enabling real-time RUL estimates accounting for aleatoric and epistemic un-certainty sources, enabling informed, risk-based maintenance operations. Bayesian filters (BFs) provide a convenient framework that meets the research goal, where the posterior probability distribution of the system state (i.e., the damage state) is recursively approximated based on a time-growing stream of CM data. Among BFs, the particle filter (PF) is possibly the best candidate, as it can solve the recursive non-linear non-Gaussian state estimation problem without requiring any local linearization or any functional approximation. Solutions are proposed to implement the PF for the DP of FRP structures undergoing fatigue degradation, developing approaches meeting the online and real-time operational requirements. Local strain measurements are employed as CM data, proving to be adequate and having practical advantages over more complex sensing techniques. DD approaches are integrated within the PF to map the relation between the system state space and the CM data, basically solving a pattern recognition problem, reducing the computational expense that structural models would pose during the PF operation. Conversely, the predictive effort is performed by damage progression models. In this regard, a novel state evolution model is postulated for cross ply laminates undergoing fatigue loading, where the interaction between damage modes (transverse cracks and induced delaminations) is acknowledged and multi-damage accumulation laws are proposed, paving the way for future works in the DP of FRP arena. The proposed framework is tested in various scenarios with increasing complexity. Results demonstrate the feasibility and potential of the proposed approach as a tool able to monitor the fatigue degradation/damage progression while simultaneously providing estimates about the remaining useful life (RUL) of composite structures.
I recenti progressi nel campo dei materiali e delle tecnologie produttive hanno rivoluzionato il campo dell'ingegneria strutturale, con l’obiettivo di realizzare strutture sempre più efficienti e ad alte prestazioni. Il cavallo di battaglia di questa rivoluzione è la categoria dei materiali compositi, la cui superiorità in termini di rigidezza e resistenza specifiche li rende particolarmente attraenti in diverse applicazioni strutturali in cui la riduzione di massa è fondamentale. Tuttavia, i materiali compositi sono particolarmente vulnerabili ai difetti che si sviluppano in fase di produzione e utilizzo e, a differenza dei metalli, sono caratterizzati da meccanismi di danneggiamento e deterioramento complessi e multivariati. Queste caratteristiche peculiari hanno impedito l’implementazione sistematica di strategie di manutenzione predittiva (PM) per strutture composite, che, nella migliore delle ipotesi, vengono ispezionate periodicamente mediante tecniche di valutazione non distruttiva (NDE). La crescente capacità di acquisire e processare dati ha accelerato lo sviluppo di strategie “data-driven” (DD), prevedendo un futuro data centrico per molte discipline scientifiche, includendo la “damage prognosis” (DP) o prognostica. Nel contesto della prognostica di strutture composite, gli approcci DD sono stati fortemente incoraggiati, poiché manca una struttura universale alla base dei modelli di deterioramento dei materiali compositi, e la disciplina soffre di interpretazioni ambigue e incoerenti. È tuttavia lecito chiedersi se il successo degli approcci DD in aree in cui non esistono modelli ed i dati sono abbondanti debba influenzare l’approccio alle discipline prognostiche, essendo questa un’area in cui i dati sono in genere scarsi e la fisica che descrive i processi è in gran parte disponibile. Inoltre, occorre considerare che le discipline prognostiche, per loro definizione, richiedono uno sforzo predittivo, il che sarebbe sufficiente ad escludere i metodi puramente DD. In questo contesto, i filtri Bayesiani (BF) si rivelano uno strumento conveniente a soddisfare gli obiettivi della ricerca, in cui la distribuzione di probabilità a posteriori dello stato del sistema (cioè lo stato del danneggiamento) viene approssimata in modo ricorsivo sulla base di un flusso di dati che misurano il comportamento della struttura. Tra i BF, il “particle filter” (PF) è forse il miglior candidato, in quanto è non richiede alcuna linearizzazione locale o approssimazione funzionale ed è in grado di trattare modelli non lineari e distribuzioni di probabilità non-Gaussiane. In questa tesi vengono proposte soluzioni per implementare il PF per la prognostica di strutture composite soggette a degrado per fatica, proponendo soluzioni che permettano di operare gli algoritmi in tempo reale e online. Misere locali di deformazione sono impiegate per monitorare la risposta della struttura, dimostrandosi adeguate e presentando vantaggi pratici rispetto a soluzioni più complesse. Gli approcci DD sono integrati all'interno del PF per sintetizzare la relazione tra lo stato del sistema e le variabili che descrivono la risposta della struttura, risolvendo sostanzialmente un problema di “pattern recognition” e riducendo il costo computazionale che i modelli strutturali introdurrebbero durante l'operazione PF. Al contrario, lo sforzo predittivo è svolto da modelli di progressione del danneggiamento. A questo proposito, viene postulato un nuovo modello di evoluzione del danneggiamento per i laminati cross-ply sottoposti a fatica, dove viene riconosciuta l'interazione tra diverse modalità di danneggiamento (microcricche e delaminazioni) e vengono proposte leggi di danneggiamento che prevedano la possibilità di avere una simultanea propagazione di diverse modalità di danneggiamento, aprendo la strada ad ulteriori avanzamenti nella prognostica dei materiali compositi. L’approccio proposto è validato in vari scenari con crescente complessità. I risultati dimostrano la fattibilità e il potenziale dell'approccio proposto come uno strumento in grado di monitorare la degradazione della fatica/progressione del danno fornendo contemporaneamente stime sulla vita utile residua delle strutture composite.
Model-based damage prognosis of composite laminated structures
Cristiani, Demetrio Luigi Maria
2021/2022
Abstract
The recent advances in materials and manufacturing technology have revolutionized the field of structural engineering, aiming for increasingly efficient and high-performance structures. The workhorse of this revolution is the category of fiber reinforced plastics (FRP), exhibiting superior capabilities in terms of specific stiffness and strength and making them particularly appealing in several structural applications where weight savings are crucial. However, FRP are notably vulnerable to flaws developing from the manufacturing process and service, and unlike metals, are featured by complex and multivariate failure and degradation modes, leading to uncertainty in the assessment of current and future material properties. These shortcomings have prevented the fielding of systematic predictive maintenance (PM) strategies for complex, large scale FRP structures, which, at best, rely on resource-in-tensive periodic inspections by non-destructive evaluation (NDE) techniques. With increasing use of FRP in several industries such as aviation, effective reliability analysis of composite structures has become imperative in recent years. This desideratum may be fulfilled by developing ad-hoc damage prognosis (DP) methods. DP methods attempt to predict the remaining time beyond which the structure is expected not to perform within desired specifications, i.e., the remaining useful life (RUL), by inferring the current state of the system (i.e., structural health monitoring), estimating its future operational environments, and then predicting its RUL. In practice, DP requires the integration of diverse fields of technology and a variety of predictive modelling approaches, including the ability to quantify the uncertainty associated with RUL predictions, to enable informed, risk-based maintenance decisions. The increased ability to sense, acquire and process data has accelerated research and development breakthroughs in the field data-driven (DD) approaches, envisaging a data-centric future for many scientific disciplines, including DP. In the context of DP of composite structures DD approaches have been strongly encouraged, as a universal structure underlying damage progression models is missing, and the field suffers from ambiguous and inconsistent interpretations. It is however questionable whether the success of DD approaches in domains where data are abundant and physics-based models do not exist should trace the route of DP, being a field where data are typically scarce and scientific knowledge defining how processes evolve is, to some extent, available. The hesitation is even more legitimated by considering that DP requires a predictive effort, and purely DD approaches are unlikely to be predictive, having intrinsic limitations on the cognitive tasks that they can perform. The goal of this research is to develop a framework for the DP of FRP structures that merges the partial knowledge of condition monitoring (CM) data from an integrated sensors network with the limited predictive abilities of damage progression models, enabling real-time RUL estimates accounting for aleatoric and epistemic un-certainty sources, enabling informed, risk-based maintenance operations. Bayesian filters (BFs) provide a convenient framework that meets the research goal, where the posterior probability distribution of the system state (i.e., the damage state) is recursively approximated based on a time-growing stream of CM data. Among BFs, the particle filter (PF) is possibly the best candidate, as it can solve the recursive non-linear non-Gaussian state estimation problem without requiring any local linearization or any functional approximation. Solutions are proposed to implement the PF for the DP of FRP structures undergoing fatigue degradation, developing approaches meeting the online and real-time operational requirements. Local strain measurements are employed as CM data, proving to be adequate and having practical advantages over more complex sensing techniques. DD approaches are integrated within the PF to map the relation between the system state space and the CM data, basically solving a pattern recognition problem, reducing the computational expense that structural models would pose during the PF operation. Conversely, the predictive effort is performed by damage progression models. In this regard, a novel state evolution model is postulated for cross ply laminates undergoing fatigue loading, where the interaction between damage modes (transverse cracks and induced delaminations) is acknowledged and multi-damage accumulation laws are proposed, paving the way for future works in the DP of FRP arena. The proposed framework is tested in various scenarios with increasing complexity. Results demonstrate the feasibility and potential of the proposed approach as a tool able to monitor the fatigue degradation/damage progression while simultaneously providing estimates about the remaining useful life (RUL) of composite structures.| File | Dimensione | Formato | |
|---|---|---|---|
|
tesiDC.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
13.43 MB
Formato
Adobe PDF
|
13.43 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/10589/184013