In mechanical, civil, and aerospace engineering, structural components inevitably degrade over time. Ensuring sustained safe operation and longevity demands diligent maintenance throughout their service life. This PhD thesis introduces advanced deep learning techniques aimed at enhancing system health monitoring, with a focus on safety-critical systems such as structural beams, frames, panels, and rotating machinery, crucial in various industrial applications. The proposed frameworks aim to address key challenges and gaps prevalent in the literature. Specifically, they concentrate on characterizing damage under changing environmental conditions, diagnosing anomalies without dependence on labelled data, enhancing comprehension of the decision-making processes within complex deep learning models to increase their transparency, and integrating existing physical knowledge of the system into the neural network training process. These methods are applied to both numerical and experimental case studies, demonstrating their effectiveness through comparisons with traditional methods and validation against experiments.
Nei campi dell' ingegneria meccanica, civile e aerospaziale, i componenti strutturali inevitabilmente si degradano nel tempo. Garantire un'operatività sicura e prolungata richiede una manutenzione diligente durante tutto il loro ciclo di vita. Questa tesi di dottorato introduce tecniche avanzate di deep learning volte a migliorare il monitoraggio della salute dei sistemi, con un focus su sistemi critici per la sicurezza come travi strutturali, telai, pannelli e macchinari rotanti, cruciali in varie applicazioni industriali. I framework proposti mirano ad affrontare sfide chiave e lacune presenti nella letteratura. In particolare, si concentrano sulla caratterizzazione dei danni in condizioni ambientali variabili, sulla diagnosi delle anomalie senza dipendere da dati etichettati, sul miglioramento della comprensione dei processi decisionali all'interno di modelli complessi di deep learning per aumentarne la trasparenza e sull'integrazione della conoscenza fisica esistente del sistema nel processo di addestramento delle reti neurali. Questi metodi vengono applicati a casi studio sia numerici che sperimentali, dimostrando la loro efficacia attraverso confronti con metodi tradizionali e validazioni con esperimenti.
Structural health monitoring and beyond : deep learning approaches for smart identification and prediction
Parziale, Marc
2023/2024
Abstract
In mechanical, civil, and aerospace engineering, structural components inevitably degrade over time. Ensuring sustained safe operation and longevity demands diligent maintenance throughout their service life. This PhD thesis introduces advanced deep learning techniques aimed at enhancing system health monitoring, with a focus on safety-critical systems such as structural beams, frames, panels, and rotating machinery, crucial in various industrial applications. The proposed frameworks aim to address key challenges and gaps prevalent in the literature. Specifically, they concentrate on characterizing damage under changing environmental conditions, diagnosing anomalies without dependence on labelled data, enhancing comprehension of the decision-making processes within complex deep learning models to increase their transparency, and integrating existing physical knowledge of the system into the neural network training process. These methods are applied to both numerical and experimental case studies, demonstrating their effectiveness through comparisons with traditional methods and validation against experiments.File | Dimensione | Formato | |
---|---|---|---|
PhD thesis - Marc Parziale.pdf
non accessibile
Descrizione: PhD thesis - Marc Parziale
Dimensione
22.85 MB
Formato
Adobe PDF
|
22.85 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/10589/221712