This paper analyzes the risk of hydrogen leakage from a watertight gasometer installed in a vegetable oil refining plant. The causes of the leak and possible accident scenarios are modeled using fault tree analysis, event tree analysis, and a Bow-Tie diagram; in the absence of historical data, the probabilities of the basic events are estimated by the judgment of a panel of experts. The Bow-Tie is then converted into a Bayesian network, which represents events common to multiple branches and enables diagnostic inference. The leak probability is 2.28 10−2; approximately two out of three leaks ignite, and a third result in a vapor cloud explosion. Inspection and operational errors are the almost certain cause of a leak, while instrumentation failures and material degradation have a negligible impact. Recalculating the model also shows that, in this plant, the Bow-Tie and Bayesian network provide virtually identical values. Finally, the results are placed within the Enterprise Risk Management perspective: a simulation of the model indicates that halving inspection errors would reduce the probability of loss by approximately half, supporting resource allocation geared toward organizational measures.
Il lavoro analizza il rischio di perdita di idrogeno da un gasometro a tenuta d’acqua installato in un impianto di raffinazione di oli vegetali. Le cause della perdita e i possibili scenari incidentali sono modellati con Fault Tree Analysis, Event Tree Analysis e diagramma Bow-Tie; in assenza di dati storici, le probabilità degli eventi base sono stimate dal giudizio di un gruppo di esperti. Il Bow-Tie è poi convertito in una rete bayesiana, che rappresenta gli eventi comuni a più rami e consente l’inferenza diagnostica. La probabilità di perdita risulta pari a 2,28·10−2; circa due perdite su tre si innescano e un terzo evolve in un’esplosione della nube di vapore. Gli errori legati alle ispezioni e all’esercizio risultano la causa quasi certa di una perdita, mentre guasti della strumentazione e degrado dei materiali hanno un peso trascurabile. Il ricalcolo del modello mostra inoltre che, in questo impianto, Bow-Tie e rete bayesiana forniscono valori praticamente coincidenti. I risultati sono infine collocati nella prospettiva dell’Enterprise Risk Management: una simulazione sul modello indica che dimezzare gli errori ispettivi ridurrebbe di circa la metà la probabilità di perdita, a sostegno di un’allocazione delle risorse orientata alle misure organizzative.
ANALISI DEL RISCHIO DI UN GASOMETRO AD IDROGENO CON BOW-TIE E RETI BAYESIANE : DALLA VALUTAZIONE TECNICA ALL’ ENTERPRISE RISK MANAGEMENT
LAURIOLA, FLAVIO
2025/2026
Abstract
This paper analyzes the risk of hydrogen leakage from a watertight gasometer installed in a vegetable oil refining plant. The causes of the leak and possible accident scenarios are modeled using fault tree analysis, event tree analysis, and a Bow-Tie diagram; in the absence of historical data, the probabilities of the basic events are estimated by the judgment of a panel of experts. The Bow-Tie is then converted into a Bayesian network, which represents events common to multiple branches and enables diagnostic inference. The leak probability is 2.28 10−2; approximately two out of three leaks ignite, and a third result in a vapor cloud explosion. Inspection and operational errors are the almost certain cause of a leak, while instrumentation failures and material degradation have a negligible impact. Recalculating the model also shows that, in this plant, the Bow-Tie and Bayesian network provide virtually identical values. Finally, the results are placed within the Enterprise Risk Management perspective: a simulation of the model indicates that halving inspection errors would reduce the probability of loss by approximately half, supporting resource allocation geared toward organizational measures.| File | Dimensione | Formato | |
|---|---|---|---|
|
LAURIOLA_TESI.pdf
File consultabile da postazioni biblioteca
Dimensione
1.36 MB
Formato
Adobe PDF
|
1.36 MB | Adobe PDF |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14253/110701