Recently, the concern for energy consumption is increasing. There is a need of new solutions towards a more sustainable manufacturing. One of the most promising measures is to control the machine energy state such that the policy does not compromise the production at system level. Machines can be switched off while starving and blocked, although a start-up is required to resume the service. The effect of machine control at system level is not straightforward due to problem complexity. For example, if only one machine is controlled, blocking and starvation effects might propagate along the production system. Therefore, it might happen that an energy saving at machine level is compromised by an increase of consumption of other machines that are more often starved or blocked. We are interested in the evaluation of the performance of transfer production lines with an energy control applied at machine level. That can be obtained using simulations or analytical methods. Through simulation, it is possible to obtain reliable results, but models are slow and expensive to build. Analytical methods, whose hypotheses do not reflect the industrial reality, are fast in execution but biased. Both these tools have big disadvantages, such that the barrier for practical application is high. Since in the industrial reality, it is necessary to obtain reliable results quickly and easily. This work develops an algorithm to evaluate the performance of a transfer line in which the state of machines is controlled for energy saving purposes. Few information, obtained from a simulation model are combined, with biased, but numerous information from analytical models. The Extended Kernel Regression method is used to obtain a meta-model based on multi-fidelity data. The low-fidelity data come from an analytical method properly selected to fit our requirement. A discrete event simulation model is created to accurately represent the system. A numerical analysis is conducted for balanced lines of five and ten machines. The results obtained for the different experimental cases were compared with the simulation model, to test the precision of the predictions. Furthermore, a comparison with simple Kernel Regression, which is based only on simulation data, is reported to evaluate the actual advantage obtained. A large set of parameters, for the line and the policy, are analyzed in order to investigate how the method works in different system conditions.
Recentemente, la preoccupazione per il consumo di energia è in aumento. C'è bisogno di nuove soluzioni per una produzione più sostenibile. Una delle misure più promettenti è controllare lo stato energetico della macchina in modo tale che la politica non comprometta la produzione a livello di sistema. Le macchine possono essere spente mentre sono in starving o blocking, anche se è necessario uno start-up per riprendere il servizio. L'effetto del controllo della macchina a livello di sistema non è immediato a causa della complessità del problema. Ad esempio, se viene controllata solo una macchina, gli effetti di blocking e starvation potrebbero propagarsi lungo il sistema di produzione. Pertanto, potrebbe accadere che un risparmio energetico a livello di macchina venga compromesso da un aumento del consumo di altre macchine che sono più spesso in starving o blocking. Siamo interessati alla valutazione delle prestazioni delle linee di produzione con un controllo energetico applicato a livello macchina. Questo può essere ottenuto usando simulazioni o metodi analitici. Attraverso la simulazione, è possibile ottenere risultati affidabili, ma i modelli sono lenti e costosi da costruire. I metodi analitici, le cui ipotesi non riflettono la realtà industriale, sono veloci nell'esecuzione ma distorti. Entrambi questi strumenti hanno grossi svantaggi, tanto che la barriera per l'applicazione pratica è elevata. Poiché nella realtà industriale è necessario ottenere risultati affidabili in modo rapido e semplice. Questo lavoro sviluppa un algoritmo per valutare le prestazioni di una linea in cui lo stato delle macchine è controllato per scopi di risparmio energetico. Poche informazioni, ottenute da un modello di simulazione sono combinate, con distorte, ma numerose informazioni dei modelli analitici. Il metodo Extended Kernel Regression viene utilizzato per ottenere un meta-modello basato su dati multi-fedelity. I dati low-fidelity provengono da un metodo analitico opportunamente selezionato per soddisfare le nostre esigenze. Viene creato un modello di simulazione di eventi discreti per rappresentare accuratamente il sistema. Un'analisi numerica è condotta per linee bilanciate di cinque e dieci macchine. I risultati ottenuti per i diversi casi sperimentali sono stati confrontati con il modello di simulazione, per testare la precisione delle previsioni. Inoltre, viene riportato un confronto con la semplice Kernel Regression, che si basa solo sui dati di simulazione, per valutare l'effettivo vantaggio ottenuto. Un'ampia serie di parametri, per la linea e la politica, vengono analizzati per indagare come il metodo funziona in condizioni di sistema diverse.
Performance evaluation of a serial production line with control for energy saving : estimation through a multi-fidelity meta-model
BUCCOLIERO, ALESSIO
2017/2018
Abstract
Recently, the concern for energy consumption is increasing. There is a need of new solutions towards a more sustainable manufacturing. One of the most promising measures is to control the machine energy state such that the policy does not compromise the production at system level. Machines can be switched off while starving and blocked, although a start-up is required to resume the service. The effect of machine control at system level is not straightforward due to problem complexity. For example, if only one machine is controlled, blocking and starvation effects might propagate along the production system. Therefore, it might happen that an energy saving at machine level is compromised by an increase of consumption of other machines that are more often starved or blocked. We are interested in the evaluation of the performance of transfer production lines with an energy control applied at machine level. That can be obtained using simulations or analytical methods. Through simulation, it is possible to obtain reliable results, but models are slow and expensive to build. Analytical methods, whose hypotheses do not reflect the industrial reality, are fast in execution but biased. Both these tools have big disadvantages, such that the barrier for practical application is high. Since in the industrial reality, it is necessary to obtain reliable results quickly and easily. This work develops an algorithm to evaluate the performance of a transfer line in which the state of machines is controlled for energy saving purposes. Few information, obtained from a simulation model are combined, with biased, but numerous information from analytical models. The Extended Kernel Regression method is used to obtain a meta-model based on multi-fidelity data. The low-fidelity data come from an analytical method properly selected to fit our requirement. A discrete event simulation model is created to accurately represent the system. A numerical analysis is conducted for balanced lines of five and ten machines. The results obtained for the different experimental cases were compared with the simulation model, to test the precision of the predictions. Furthermore, a comparison with simple Kernel Regression, which is based only on simulation data, is reported to evaluate the actual advantage obtained. A large set of parameters, for the line and the policy, are analyzed in order to investigate how the method works in different system conditions.| File | Dimensione | Formato | |
|---|---|---|---|
|
AB_MasterThesis.pdf
non accessibile
Descrizione: Master Thesis - Alessio Buccoliero
Dimensione
2.63 MB
Formato
Adobe PDF
|
2.63 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/10589/146605