The objective of this thesis is to develop a fault diagnosis and prognosis system, implemented for a Brushless DC (BLDC) motor, with the intent to extend it to any complex dynamic system for which there is no in-depth knowledge of the mathematical model. The adopted approach is hybrid, integrating structural analysis techniques, parameter estimation methods, and degradation models, with the aim of providing an accurate assessment of the system's Remaining Useful Life (RUL). A central element of this work has been the use of an existing toolbox from the literature, which allowed the extraction of the structural model of the BLDC motor and the determination of the MSO (Minimal Structurally Overdetermined) sets. These sets enabled the identification of redundant constraints within the system, which were subsequently used to generate residual signals, fundamental for fault diagnosis. These residuals represent a key element for monitoring the health status of the motor, facilitating the timely detection of potential anomalies. In the context of fault diagnosis, one of the main challenges addressed was minimizing false alarms, often caused by the presence of noise in the residual signals. To tackle this issue, the CUSUM algorithm was employed, demonstrating its effectiveness in providing a robust method for diagnosis, even in real-time applications. Thanks to this algorithm, it was possible to constantly monitor the residuals and promptly identify any violations of diagnostic thresholds, significantly reducing the risk of false alarms. Regarding fault prognosis, the results obtained through parameter estimation algorithms, such as RLS (Recursive Least Squares) and ZRLS (Zonotopic Recursive Least Squares), demonstrated the system's capability to deliver accurate predictions of the motor's RUL. Both approaches, despite being based on different theories, produced similar results in terms of accuracy and reliability. However, the ZRLS algorithm exhibited greater robustness in handling noise, enabling the calculation of confidence intervals and providing estimates of the uncertainty associated with RUL predictions. Moreover, the use of degradation models allowed the quantification of fault evolution over time, offering critical information for planning maintenance interventions. The hybrid approach adopted in this thesis has proven, despite the limited amount of synthetic data, to be effective in both fault diagnosis and RUL prediction. The results obtained have the potential to be applied in real industrial scenarios, providing a versatile model for monitoring the operating conditions of complex systems. The implications of this work extend beyond the BLDC motor, suggesting that similar methodologies could be implemented across various industrial sectors.
Il presente lavoro di tesi ha come obiettivo lo sviluppo di un sistema di diagnosi e prognosi guasti, implementato per un motore Brushless DC (BLDC), ma con l'intento di estenderlo a qualsiasi sistema dinamico complesso di cui non si ha una conoscenza approfondita del modello matematico. L’approccio adottato è di tipo ibrido e integra tecniche di analisi strutturale, metodi di stima parametrica e modelli di degradazione, con l'obiettivo di fornire una valutazione accurata della vita utile residua (RUL - Remaining Useful Life) del sistema. Uno degli elementi centrali di questo lavoro è stato l’utilizzo di un toolbox già presente in letteratura, che ha consentito di ricavare il modello strutturale del motore BLDC e di determinare gli insiemi MSO (Minimal Structurally Overdetermined). Questi insiemi hanno permesso di identificare i vincoli ridondanti all'interno del sistema, successivamente utilizzati per generare i segnali residui, fondamentali per la diagnosi dei guasti. Tali residui costituiscono un elemento chiave per il monitoraggio dello stato di salute del motore, facilitando l’individuazione tempestiva di eventuali anomalie. Nel contesto della diagnosi guasti, uno dei principali problemi affrontati è stato quello di minimizzare i falsi allarmi, spesso generati dalla presenza di rumore nei segnali residui. Per far fronte a questa problematica, è stato utilizzato l’algoritmo CUSUM, che ha dimostrato la sua efficacia nel fornire un metodo robusto di diagnosi applicabile anche in tempo reale. Grazie a questo algoritmo, è stato possibile monitorare costantemente i residui e identificare in modo tempestivo eventuali violazioni delle soglie diagnostiche, riducendo al minimo il rischio di falsi allarmi. Per quanto riguarda la prognosi dei guasti, i risultati ottenuti attraverso gli algoritmi di stima parametrica, come RLS (Recursive Least Squares) e ZRLS (Zonotopic Recursive Least Squares), hanno evidenziato la capacità del sistema di fornire previsioni accurate sulla vita utile residua del motore. Entrambi gli approcci, pur basandosi su teorie differenti, hanno prodotto risultati simili in termini di precisione e affidabilità. Tuttavia, l’algoritmo ZRLS ha dimostrato una maggiore robustezza nella gestione del rumore, permettendo di calcolare intervalli di confidenza e di stimare l’incertezza associata alle previsioni di RUL. Inoltre, l’uso di modelli di degradazione ha consentito di quantificare l’evoluzione del guasto nel tempo, offrendo informazioni per la pianificazione di eventuali interventi di manutenzione. L’approccio ibrido adottato in questa tesi ha dimostrato, seppur con una quantità limitata di dati sintetici, di essere efficace sia nella diagnosi dei guasti sia nella previsione della RUL. I risultati ottenuti possono essere applicabili in scenari industriali reali, fornendo un modello versatile per il monitoraggio delle condizioni operative di sistemi complessi. Le implicazioni di questo lavoro vanno oltre il motore BLDC, suggerendo metodologie simili che potrebbero essere implementate in vari settori industriali.
Studio e Implementazione di un Metodo per la Diagnosi e la Prognosi Guasti tramite la Combinazione di Analisi Strutturale e Tecniche Data-Driven
SERAFINI, ANDREA
2023/2024
Abstract
The objective of this thesis is to develop a fault diagnosis and prognosis system, implemented for a Brushless DC (BLDC) motor, with the intent to extend it to any complex dynamic system for which there is no in-depth knowledge of the mathematical model. The adopted approach is hybrid, integrating structural analysis techniques, parameter estimation methods, and degradation models, with the aim of providing an accurate assessment of the system's Remaining Useful Life (RUL). A central element of this work has been the use of an existing toolbox from the literature, which allowed the extraction of the structural model of the BLDC motor and the determination of the MSO (Minimal Structurally Overdetermined) sets. These sets enabled the identification of redundant constraints within the system, which were subsequently used to generate residual signals, fundamental for fault diagnosis. These residuals represent a key element for monitoring the health status of the motor, facilitating the timely detection of potential anomalies. In the context of fault diagnosis, one of the main challenges addressed was minimizing false alarms, often caused by the presence of noise in the residual signals. To tackle this issue, the CUSUM algorithm was employed, demonstrating its effectiveness in providing a robust method for diagnosis, even in real-time applications. Thanks to this algorithm, it was possible to constantly monitor the residuals and promptly identify any violations of diagnostic thresholds, significantly reducing the risk of false alarms. Regarding fault prognosis, the results obtained through parameter estimation algorithms, such as RLS (Recursive Least Squares) and ZRLS (Zonotopic Recursive Least Squares), demonstrated the system's capability to deliver accurate predictions of the motor's RUL. Both approaches, despite being based on different theories, produced similar results in terms of accuracy and reliability. However, the ZRLS algorithm exhibited greater robustness in handling noise, enabling the calculation of confidence intervals and providing estimates of the uncertainty associated with RUL predictions. Moreover, the use of degradation models allowed the quantification of fault evolution over time, offering critical information for planning maintenance interventions. The hybrid approach adopted in this thesis has proven, despite the limited amount of synthetic data, to be effective in both fault diagnosis and RUL prediction. The results obtained have the potential to be applied in real industrial scenarios, providing a versatile model for monitoring the operating conditions of complex systems. The implications of this work extend beyond the BLDC motor, suggesting that similar methodologies could be implemented across various industrial sectors.File | Dimensione | Formato | |
---|---|---|---|
Andrea_Serafini_Tesi.pdf
accesso aperto
Dimensione
7.04 MB
Formato
Adobe PDF
|
7.04 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.12075/19226