Industrial production is currently experiencing a fast-paced digital transformation, supported by initiatives within the technological framework of Industry 4.0. Thanks to the acquisition of diverse sets of data and their quantitative analysis aimed at discovering useful insights, factories are becoming more automated and efficient. This thesis belongs to this context and draws its motivation from a project sponsored by Regione Lombardia and jointly developed by Politecnico di Milano and Pirelli Tyre S.p.A., in which the goal is to increase the efficiency of tire production leveraging Machine-Learning techniques to analyze data coming from a highly automated facility. This work lays the foundations of the project’s development, identifying and analyzing the central issues of the production process, of which the most relevant one is the impact of the long sequence of sub-processes performed in the production line on the quality of final products. We analyze the problem by decomposing it into two different sub-problems: Defect Isolation, i.e., the identification of root causes of defects, and Fault Prediction, i.e., the estimation of the possible faulty behavior of machines in the near-future. We propose methods to tackle these problems and we test them both on a first set of real data coming from the industrial plant under study, and, whenever data are not yet available, on a simulated environment that was specifically designed to resemble the features of the real production plant. After a discussion on the results obtained so far with either real or simulated data, we describe what are the possible useful extensions of the proposed methodology and the future steps in the industrial project’s course.
La produzione industriale attuale sta subendo un rapido processo di trasformazione digitale, supportato da iniziative all’interno del contesto Industria 4.0. Grazie alla raccolta di diversi insiemi di dati ed alla loro analisi quantitativa e mirata ad estrarre informazioni utili, le fabbriche stanno via via diventando più automatizzate ed efficienti. Questa tesi appartiene a questo contesto, e ha le sue motivazioni in un progetto supportato da Regione Lombardia e congiuntamente sviluppato da Politecnico di Milano e Pirelli Tyre S.p.a., in cui l’obiettivo è quello di aumentare l’efficienza della produzione di pneumatici sfruttando tecniche di Machine Learning per analizzare dati provenienti da un impianto ad alta automatizzazione. Il presente lavoro pone le basi per lo sviluppo del progetto, identificando e analizzando i problemi centrali del processo di produzione in questione, di cui il più rilevante è l’impatto della lunga sequenza di sottoprocessi che vengono effettuati sulla qualità finale dello pneumatico. Analizziamo questo problema decomponendolo in due different sotto-problemi: Defect Isolation, i.e., l’identificazione delle cause alla radice dei difetti, e Fault Prediction, i.e., la predizione di possibili guasti delle macchine in un futuro prossimo. Di seguito proponiamo un insieme di metodi per affrontare questi problemi, testandoli su dati provenienti dalla linea di produzione sotto esame, quando questi sono disponibili, o su un ambiente simulato creato appositamente per incapsulare proprietà interessanti del caso reale. Al termine di questa discussione, analizzieremo i risultati ottenuti con dati sintetici o reali, descrivendo le possibili evoluzioni ed estensioni nell’ambito del progetto.
Machine Learning approaches to increase production efficiency : an Industry 4.0 case
MONTALI, NICO
2017/2018
Abstract
Industrial production is currently experiencing a fast-paced digital transformation, supported by initiatives within the technological framework of Industry 4.0. Thanks to the acquisition of diverse sets of data and their quantitative analysis aimed at discovering useful insights, factories are becoming more automated and efficient. This thesis belongs to this context and draws its motivation from a project sponsored by Regione Lombardia and jointly developed by Politecnico di Milano and Pirelli Tyre S.p.A., in which the goal is to increase the efficiency of tire production leveraging Machine-Learning techniques to analyze data coming from a highly automated facility. This work lays the foundations of the project’s development, identifying and analyzing the central issues of the production process, of which the most relevant one is the impact of the long sequence of sub-processes performed in the production line on the quality of final products. We analyze the problem by decomposing it into two different sub-problems: Defect Isolation, i.e., the identification of root causes of defects, and Fault Prediction, i.e., the estimation of the possible faulty behavior of machines in the near-future. We propose methods to tackle these problems and we test them both on a first set of real data coming from the industrial plant under study, and, whenever data are not yet available, on a simulated environment that was specifically designed to resemble the features of the real production plant. After a discussion on the results obtained so far with either real or simulated data, we describe what are the possible useful extensions of the proposed methodology and the future steps in the industrial project’s course.| File | Dimensione | Formato | |
|---|---|---|---|
|
2018_04_Montali.pdf
non accessibile
Descrizione: Thesis text
Dimensione
3.29 MB
Formato
Adobe PDF
|
3.29 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/10589/140293