Safety critical energy systems (such as Nuclear Power Plants (NPPs), chemical and Oil & Gas facilities…) rely on safety barriers to avoid that accidents occur (preventive barriers) and to mitigate the consequences if they were to occur (mitigative barriers). However, safety barriers degrade in time, eventually leading to a larger risk of severe consequences. The typical approaches proposed in the literature for the assessment of safety barriers performance neglect the influence of such degradation, which, instead, has a great influence on the safety response of the system during an accident. To address these limitations, this PhD thesis develops methods for the risk assessment and management of degrading energy systems, supporting both operational and design decisions. The main contributions of this work are: • A Bayesian Network (BN)-based framework for the multi-hazard risk assessment of Oil & Gas (O&G) plants, which originally integrates time dependent Key Performance Indicators (KPIs) to model safety barriers degradation, a Bayesian Regularized Artificial Neural Network (BRANN) for the ignition probability estimation following a Loss Of Primary Containment (LOPC) and a multi-hazard and multi-source risk aggregation procedure to calculate the Location Specific Individual Risk (LSIR) accounting for the uncertainties on the frequency and severity of the accidental scenarios. • A risk-based maintenance optimization framework, which originally embeds fragility surfaces that characterize the safety barriers time-dependent fragility to hazards within a Dynamic Bayesian Network (DBN) model for the risk assessment. The DBN is used as the environment within a Deep Reinforcement Learning (DRL) algorithm to find the optimal safety barrier maintenance policy. • A risk-informed design framework for multi-unit systems, which combines global sensitivity indices based on Optimal Transport (OT) with a graphical visualization tool based on CUmulative SUm of NOrmalized Reordered Output (CUSUNORO). The OT-based indices allow the identification of the most important inputs, also in presence of multivariate and time-dependent inputs, whereas the CUSUNORO enables the identification of the safety critical ranges of these inputs, that ultimately enable targeted design improvements to keep the system far from the failure region. The proposed methods have been successfully applied to realistic energy systems, namely NPPs, O&G plants and chemical storage facilities.
I sistemi energetici critici (come le Centrali Nucleari (NPPs), gli impianti chimici e Oil & Gas…) sono equipaggiati con barriere di sicurezza per evitare il verificarsi di incidenti (barriere preventive) e per mitigarne le conseguenze qualora essi avvengano (barriere mitigative). Tuttavia, le barriere di sicurezza si degradano nel tempo, portando ad un aumento del rischio di conseguenze gravi. Gli approcci tipicamente proposti in letteratura per la valutazione delle prestazioni delle barriere di sicurezza trascurano l’influenza del degrado, che, invece, ha un impatto significativo sulla risposta di sicurezza del sistema durante un incidente. Per superare queste limitazioni, questa tesi di dottorato sviluppa metodi per la valutazione e la gestione del rischio di sistemi energetici soggetti a degradazione, a supporto sia delle decisioni operative sia di quelle di progetto. I principali contributi di questo lavoro sono: • Un framework basato su Reti Bayesiane (Bayesian Networks, BN) per la valutazione del rischio multi-hazard in impianti Oil & Gas (O&G), che integra in modo originale Key Performance Indicators (KPI) dipendenti dal tempo per modellare la degradazione delle barriere di sicurezza, una Bayesian Regularized Artificial Neural Network (BRANN) per la stima della probabilità di innesco a seguito di un Loss Of Primary Containment (LOPC) e una procedura di aggregazione del rischio multi-hazard e multi-sorgente per il calcolo del Location Specific Individual Risk (LSIR), tenendo conto delle incertezze sulla frequenza e sulla severità degli scenari incidentali. • Un framework di ottimizzazione della manutenzione, che utilizza in modo originale superfici di fragilità per caratterizzare la fragilità tempo-dipendente delle barriere di sicurezza rispetto ai pericoli all’interno di una Dynamic Bayesian Network (DBN) per la valutazione del rischio. La DBN è poi utilizzata come ambiente all’interno di un algoritmo di Deep Reinforcement Learning (DRL) per individuare la politica ottimale di manutenzione delle barriere di sicurezza. • Un framework di progettazione risk-informed per sistemi multi-unità, che combina indici di sensibilità globale basati sulla teoria dell’Optimal Transport (OT) con uno strumento di visualizzazione grafica basato sul CUmulative SUm of NOrmalized Reordered Output (CUSUNORO). Gli indici basati su OT consentono l’identificazione degli input più importanti anche in presenza di input multivariati e tempo-dipendenti, mentre il CUSUNORO permette di individuare gli intervalli di valori di tali input critici per la sicurezza. I metodi proposti sono stati applicati con successo a sistemi energetici realistici, quali centrali nucleari (NPPs), impianti Oil & Gas e impianti di stoccaggio di sostanze chimiche.
Advanced methods for the risk assessment and management of degrading energy systems
MARCHETTI, STEFANO
2026
Abstract
Safety critical energy systems (such as Nuclear Power Plants (NPPs), chemical and Oil & Gas facilities…) rely on safety barriers to avoid that accidents occur (preventive barriers) and to mitigate the consequences if they were to occur (mitigative barriers). However, safety barriers degrade in time, eventually leading to a larger risk of severe consequences. The typical approaches proposed in the literature for the assessment of safety barriers performance neglect the influence of such degradation, which, instead, has a great influence on the safety response of the system during an accident. To address these limitations, this PhD thesis develops methods for the risk assessment and management of degrading energy systems, supporting both operational and design decisions. The main contributions of this work are: • A Bayesian Network (BN)-based framework for the multi-hazard risk assessment of Oil & Gas (O&G) plants, which originally integrates time dependent Key Performance Indicators (KPIs) to model safety barriers degradation, a Bayesian Regularized Artificial Neural Network (BRANN) for the ignition probability estimation following a Loss Of Primary Containment (LOPC) and a multi-hazard and multi-source risk aggregation procedure to calculate the Location Specific Individual Risk (LSIR) accounting for the uncertainties on the frequency and severity of the accidental scenarios. • A risk-based maintenance optimization framework, which originally embeds fragility surfaces that characterize the safety barriers time-dependent fragility to hazards within a Dynamic Bayesian Network (DBN) model for the risk assessment. The DBN is used as the environment within a Deep Reinforcement Learning (DRL) algorithm to find the optimal safety barrier maintenance policy. • A risk-informed design framework for multi-unit systems, which combines global sensitivity indices based on Optimal Transport (OT) with a graphical visualization tool based on CUmulative SUm of NOrmalized Reordered Output (CUSUNORO). The OT-based indices allow the identification of the most important inputs, also in presence of multivariate and time-dependent inputs, whereas the CUSUNORO enables the identification of the safety critical ranges of these inputs, that ultimately enable targeted design improvements to keep the system far from the failure region. The proposed methods have been successfully applied to realistic energy systems, namely NPPs, O&G plants and chemical storage facilities.| File | Dimensione | Formato | |
|---|---|---|---|
|
Stefano_Marchetti_PhD_thesis.pdf
accesso aperto
Licenza:
Tutti i diritti riservati
Dimensione
8.14 MB
Formato
Adobe PDF
|
8.14 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/20.500.14242/384637
URN:NBN:IT:POLIMI-384637