In this thesis work, the multi-state modelling of the oxidation degradation process affecting the Gas turbine nozzles is investigated under two different maintenance settings: Condition Based Maintenance (CBM) and Predictive Maintenance (PrM). In details, the mechanism is modelled as a four-state semi-Markov process, where the possible transitions are those from a state to the next degraded state, only, and the transition times obey Weibull probability distributions. With respect to the CBM strategy, developed in the first part of the thesis work, an approach based on the Maximum Likelihood Estimation technique is proposed to characterize the parameters of the semi-Markov model, whereas the uncertainty in these values is estimated by implementing a bootstrap technique for the estimation of Fisher Information Matrix. Then, the degradation model, with uncertain parameters, is embedded in a maintenance model, with the aim of optimizing the CBM approach applied to the turbine with respect to two conflicting objective functions: unavailability and cost. The introduction of the uncertainty in the model parameters entails that the fitness values are uncertain; this requires developing a method to solve multi-optimization problems in the presence of uncertainty. Then, two techniques of the literature based on the Multi Objective Genetic Algorithms are proposed to optimize the CBM approach: NSGA-II and a combination of NSGA-II and CVaR index. With regard to PrM policy, developed in the second part of this thesis work, the problem of the estimation of the parameters of multi-state Hidden Semi-Markov models is tackled. In this case, the actual degradation state cannot be directly known, and the observation data come from a classifier that acquires signals from the monitored physical data, and infers the degradation sate. An iterative algorithm based on the Expectation-Maximization technique is developed to estimate the Hidden semi-Markov model parameters and the probabilities of miss-classification.
In questa tesi è stato trattato il problema della stima dei parametri in un modello di degrado multi stato in caso di politica manutentiva di tipo Condition Base Maintenance (CBM) e Predictive Maintenance (PrM). In particolare, il primo caso è stato modellizzato con una struttura semi-Markoviana, i cui tassi di transizione sono supposti seguire una distribuzione Weibull. La stima di questi parametri si basa su un approccio numerico della tecnica di massima verosimiglianza mentre l’incertezza di queste stime è stata caratterizzata attraverso la Fisher Information Matrix. Questi parametri incerti sono stati successivamente introdotti in un modello manutentivo con lo scopo di minimizzare i valori di indisponibilità e costo dell’impianto industriale in considerazione. L’ottimizzazione viene eseguita con tre differenti approcci basati sugli algoritmi genetici. Infine due algoritmi di decision-making sono stati presentati per aiutare l’utente alla scelta finale. Rispetto alla politica PrM, nella seconda parte di questo lavoro è stato proposto un modello semi-Markoviano nascosto in cui la stima dello stato di salute del componente è effettuata da un classificatore che monitora costantemente le variabili fisiche del sistema. Si propone un algoritmo iterativo di Expectation Maximization (EM) per la stima dei tassi di transizione del modello nascosto e delle componenti della matrice che definiscono le probabilità di misclassificazione del classificatore.
Multistate degradation semi Markov modeling and maintenance optimization
MARTINI, FABIO
2013/2014
Abstract
In this thesis work, the multi-state modelling of the oxidation degradation process affecting the Gas turbine nozzles is investigated under two different maintenance settings: Condition Based Maintenance (CBM) and Predictive Maintenance (PrM). In details, the mechanism is modelled as a four-state semi-Markov process, where the possible transitions are those from a state to the next degraded state, only, and the transition times obey Weibull probability distributions. With respect to the CBM strategy, developed in the first part of the thesis work, an approach based on the Maximum Likelihood Estimation technique is proposed to characterize the parameters of the semi-Markov model, whereas the uncertainty in these values is estimated by implementing a bootstrap technique for the estimation of Fisher Information Matrix. Then, the degradation model, with uncertain parameters, is embedded in a maintenance model, with the aim of optimizing the CBM approach applied to the turbine with respect to two conflicting objective functions: unavailability and cost. The introduction of the uncertainty in the model parameters entails that the fitness values are uncertain; this requires developing a method to solve multi-optimization problems in the presence of uncertainty. Then, two techniques of the literature based on the Multi Objective Genetic Algorithms are proposed to optimize the CBM approach: NSGA-II and a combination of NSGA-II and CVaR index. With regard to PrM policy, developed in the second part of this thesis work, the problem of the estimation of the parameters of multi-state Hidden Semi-Markov models is tackled. In this case, the actual degradation state cannot be directly known, and the observation data come from a classifier that acquires signals from the monitored physical data, and infers the degradation sate. An iterative algorithm based on the Expectation-Maximization technique is developed to estimate the Hidden semi-Markov model parameters and the probabilities of miss-classification.File | Dimensione | Formato | |
---|---|---|---|
2014_07_MARTINI.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
1.83 MB
Formato
Adobe PDF
|
1.83 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/10589/94583