As engineering systems grow more complex, advanced monitoring and diagnostic strategies are vital to ensure safety, reliability, and efficiency—especially in aviation, where components are subject to extreme loading. Traditional structural damage assessment methods, though generally reliable, are limited by high costs, time inefficiencies, and unsuitability for real-time detection. These shortcomings highlight the need for predictive tools to enable continuous performance monitoring and cost-effective maintenance while enhancing mission reliability and safety. This dissertation explores Structural Health Monitoring (SHM) with a focus on vibration-based methods, which enable real-time damage detection and predictive maintenance. These techniques, supported by affordable sensing technologies and wireless data acquisition, offer a viable alternative to traditional strategies. However, their effectiveness relies on advanced data processing capable of identifying condition-sensitive damage indicators in complex and noisy environments. The research develops machine learning (ML)-based SHM frameworks for rotary machinery and transmission systems operating under harsh conditions. It emphasizes real-time vibration monitoring and fault detection in rotary shafts, including cases of operational and ballistic impact damage. Data-driven and physics-informed ML techniques—including CNNs, LSTMs, and NARX models—are employed for signal analysis and fault classification. A key contribution is the integration of Physics-Informed Neural Networks (PINNs) within surrogate models to enhance prediction accuracy, physical consistency, and generalization. These approaches offer robust, scalable, and computationally efficient solutions for fault diagnosis. The outcomes lay a strong foundation for intelligent SHM systems, advancing safety, reliability, and performance in aerospace applications.
Con l’aumentare della complessità dei sistemi ingegneristici, strategie avanzate di monitoraggio e diagnostica diventano fondamentali per garantire sicurezza, affidabilità ed efficienza, soprattutto nel settore aeronautico, dove i componenti sono sottoposti a condizioni di carico estreme. I metodi tradizionali di valutazione dei danni strutturali, sebbene generalmente affidabili, presentano limiti legati ad alti costi, inefficienze temporali e all’impossibilità di rilevamento in tempo reale. Tali criticità evidenziano la necessità di strumenti predittivi in grado di monitorare continuamente le prestazioni e di ottimizzare la manutenzione, migliorando l’affidabilità della missione e la sicurezza dell’equipaggio. Questa tesi esplora l’utilizzo dello Structural Health Monitoring (SHM), con particolare attenzione ai metodi basati sulle vibrazioni, che permettono il rilevamento dei danni in tempo reale e la manutenzione predittiva. Supportate dalla disponibilità di tecnologie di sensorizzazione a basso costo e trasmissione wireless, queste tecniche offrono un’alternativa efficace rispetto agli approcci tradizionali. Tuttavia, la loro efficacia dipende da avanzati processi di analisi dei dati in grado di identificare indicatori di danno in ambienti complessi e rumorosi. La ricerca sviluppa framework SHM basati su tecniche di machine learning (ML) per macchinari rotanti e sistemi di trasmissione operanti in condizioni ambientali severe, con particolare attenzione al monitoraggio in tempo reale e all’identificazione di guasti, inclusi danni da impatto balistico. Le tecniche ML, tra cui CNN, LSTM e modelli NARX, sono integrate con Physics-Informed Neural Networks (PINN), migliorando accuratezza predittiva, coerenza fisica e capacità di generalizzazione. I risultati offrono soluzioni robuste e scalabili per la diagnostica e la manutenzione predittiva, contribuendo all’evoluzione di sistemi SHM intelligenti per applicazioni aerospaziali.
AI-driven structural health monitoring for rotary machinery in aerospace applications
Panagiotopoulou, Vasiliki
2024/2025
Abstract
As engineering systems grow more complex, advanced monitoring and diagnostic strategies are vital to ensure safety, reliability, and efficiency—especially in aviation, where components are subject to extreme loading. Traditional structural damage assessment methods, though generally reliable, are limited by high costs, time inefficiencies, and unsuitability for real-time detection. These shortcomings highlight the need for predictive tools to enable continuous performance monitoring and cost-effective maintenance while enhancing mission reliability and safety. This dissertation explores Structural Health Monitoring (SHM) with a focus on vibration-based methods, which enable real-time damage detection and predictive maintenance. These techniques, supported by affordable sensing technologies and wireless data acquisition, offer a viable alternative to traditional strategies. However, their effectiveness relies on advanced data processing capable of identifying condition-sensitive damage indicators in complex and noisy environments. The research develops machine learning (ML)-based SHM frameworks for rotary machinery and transmission systems operating under harsh conditions. It emphasizes real-time vibration monitoring and fault detection in rotary shafts, including cases of operational and ballistic impact damage. Data-driven and physics-informed ML techniques—including CNNs, LSTMs, and NARX models—are employed for signal analysis and fault classification. A key contribution is the integration of Physics-Informed Neural Networks (PINNs) within surrogate models to enhance prediction accuracy, physical consistency, and generalization. These approaches offer robust, scalable, and computationally efficient solutions for fault diagnosis. The outcomes lay a strong foundation for intelligent SHM systems, advancing safety, reliability, and performance in aerospace applications.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_05_Panagiotopoulou_Vasiliki.pdf
accessibile in internet per tutti a partire dal 27/05/2028
Descrizione: PhD thesis
Dimensione
40.37 MB
Formato
Adobe PDF
|
40.37 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/10589/238817