The manufacturing industry is increasingly challenged by the growing demand to control and reduce its energy consumption. In this context, the wood manufacturing industry also shows interest, facing an environment of growing competitiveness and sustainability-related policies. Complex woodworking centres are under study to evaluate their energy performance. Nowadays, these work centres are equipped with sensors to collect data on power and activity of the individual machine components. Industrial interest is focused on the development of data-driven tools capable of automatically assessing the efficiency of machine tools during production phases. Online monitoring allows for the visualization and storage of measured data; however, further analysis is needed to draw conclusions about the machine’s energy behaviour and its potential performance in the near future. Leveraging information on the activity status of the different machine components could be a starting point for better characterizing and understanding the data collected. This thesis first investigates the issue of identifying different energy states and then focuses on the problem of energy consumption prediction. The proposed methodology for the first problem involves an unsupervised Machine Learning (ML) algorithm known as K means, where the input data is provided by monitoring and consists of information regarding active power consumption and the activity status of various components. To address the second problem, two different methods are developed: one based on empirical distribution sampling and one employing a supervised ML algorithm known as Long Short-Term Memory. For numerical analyses, real data acquired from an industrial environment are used. Proper training of ML algorithms is necessary, and guidelines for the industrial application of both methodologies are provided in the conclusion of this work.
L’industria manufatturiera è sempre più sfidata da una crescente richiesta di controllare e ridurre il proprio impatto in termini di consumo energetico. In questo ambito si colloca anche l’interesse dell’industria manufatturiera del legno, in un contesto di crescente competitività e politiche relative alla sostenibilità. Complessi centri di lavoro per la lavorazione del legno sono oggetti di studio per valutarne il comportamento energetico. Oggigiorno, i centri di lavoro sono equipaggiati con sensori per raccogliere dati relativi alla potenza ed all’attività dei singoli componenti del macchinario. L’interesse industriale è focalizzato sullo sviluppo di strumenti data driven in grado di valutare automaticamente l’efficienza delle machine utensili durante le fasi di produzione. Il monitoraggio on line permette di visualizzare ed immagazzinare i dati misurati, tuttavia, ulteriori analisi sono richieste per trarre delle conclusioni sul comportamento energetico della macchina e sul come potrebbe comportarsi nell’immediato futuro. Utilizzare le informazioni relative allo stato di attività dei differenti componenti della macchina potrebbe essere un punto di partenza per poter caratterizzare e comprendere meglio i dati raccolti. Questa tesi investiga prima il problema dell’identificazione dei diversi stati energetici e successivamente si focalizza sul problema della predizione del consumo energetico. La metodologia proposta per il primo problema coinvolge un algoritmo di Machine Learning (ML) non supervisionato conosciuto come K-means, dove i dati in input sono forniti dal monitoraggio e consistono nelle informazioni riguardanti il consumo di potenza attiva e lo stato di attività dei vari componenti. Per fronteggiare il secondo problema vengono sviluppati due diversi metodi, uno basato sul campionamento di distribuzioni empiriche ed uno facente uso di un algoritmo supervisionato di ML noto come Long Short-Term Memory. Per le analisi numeriche vengono usati dati reali, acquisiti da un ambiente industriale. Un adeguato addestramento degli algoritmi di ML è necessario e delle linee guida per l’applicazione industriale di entrambe le metodologie sono riportati nella conclusione di questo lavoro.
Energy consumption in woodcutting industry: data-driven approach for energy classification and prediction
MARCONI, SEBASTIANO
2023/2024
Abstract
The manufacturing industry is increasingly challenged by the growing demand to control and reduce its energy consumption. In this context, the wood manufacturing industry also shows interest, facing an environment of growing competitiveness and sustainability-related policies. Complex woodworking centres are under study to evaluate their energy performance. Nowadays, these work centres are equipped with sensors to collect data on power and activity of the individual machine components. Industrial interest is focused on the development of data-driven tools capable of automatically assessing the efficiency of machine tools during production phases. Online monitoring allows for the visualization and storage of measured data; however, further analysis is needed to draw conclusions about the machine’s energy behaviour and its potential performance in the near future. Leveraging information on the activity status of the different machine components could be a starting point for better characterizing and understanding the data collected. This thesis first investigates the issue of identifying different energy states and then focuses on the problem of energy consumption prediction. The proposed methodology for the first problem involves an unsupervised Machine Learning (ML) algorithm known as K means, where the input data is provided by monitoring and consists of information regarding active power consumption and the activity status of various components. To address the second problem, two different methods are developed: one based on empirical distribution sampling and one employing a supervised ML algorithm known as Long Short-Term Memory. For numerical analyses, real data acquired from an industrial environment are used. Proper training of ML algorithms is necessary, and guidelines for the industrial application of both methodologies are provided in the conclusion of this work.File | Dimensione | Formato | |
---|---|---|---|
2024_12_MARCONI_Tesi.pdf
solo utenti autorizzati a partire dal 14/11/2027
Descrizione: Tesi
Dimensione
4.2 MB
Formato
Adobe PDF
|
4.2 MB | Adobe PDF | Visualizza/Apri |
2024_12_Marconi_Executive summary.pdf
solo utenti autorizzati a partire dal 14/11/2027
Descrizione: Executive summary
Dimensione
925.27 kB
Formato
Adobe PDF
|
925.27 kB | 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/10589/230579