Nowadays, manufacturing companies are witnessing a high level of customization required by customers and an increase in competition from low-cost countries. The evolution of the competitive context has prompted companies to compete not only on product price and quality, but also on the service offered to the customer. In fact, companies try to offer customized products and to deliver quickly in compliance with the delivery date negotiated with the customer. In order meet these market demands, companies adopt a Make-to-Order policy, which means that the products are only manufactured from the moment the customer's order arrives. Production planning is therefore a crucial factor for the success of the company. One of the most discussed techniques to deal with this situation is the Workload Control. Workload Control is an order release management tool that aims to keep the workload in the system stable and limited in order to reduce shop congestion and get products flowing quickly through the production process. Workload Control literature is very abundant. In recent decades, many models have been proposed with different characteristics and objectives. However, a dominant model has not been identified since the performance of the models depends on numerous factors such as the configuration of the production system, the parameters of the model itself, the parameters describing the demand, etc. In a Make-to-Order context characterized by uncertain and variable demand, it may be interesting from a practical point of view understanding which model is most suitable according to the characteristics of the demand. A mathematical tool to support this decision can be the use of Machine Learning models that can be trained to associate the best model with a set of variables that describe the characteristics¬ of the demand. The goal of this thesis is to test different classification techniques that identify the order release model that returns the best performance in terms of average Gross Throughput Time.
Al giorno d'oggi si sta assistendo a un alto livello di personalizzazione richiesto dai clienti e a un aumento della competizione dai paesi a basso costo. L'evoluzione del contesto competitivo ha spinto le aziende manifatturiere a competere non solo su prezzo e qualità del prodotto, ma anche sul servizio offerto al cliente. Le aziende cercano infatti di offrire prodotti customizzati e di consegnare rapidamente nel rispetto della data di consegna negoziata col cliente. Per soddisfare queste richieste del mercato, viene adottata una politica Make-to-Order, il che significa che i prodotti vengono realizzati soltanto dal momento in cui giunge l’ordine del cliente. La pianificazione della produzione risulta dunque un fattore cruciale per il successo della azienda. Una delle tecniche più discusse per fronteggiare questa situazione è il Workload Control. Il Workload Control è uno strumento di gestione del rilascio degli ordini in produzione che ha l'obiettivo di mantenere stabile e limitato il carico di lavoro nel sistema per ridurre la congestione dello shop e far fluire rapidamente i prodotti attraverso il processo di produzione. La letteratura del Workload Control è molto ricca. Negli ultimi decenni sono stati proposti molti modelli con caratteristiche e obiettivi diversi. Tuttavia, non è stato identificato un modello dominante poiché le prestazioni dei modelli dipendono da numerosi fattori come la configurazione del sistema di produzione, i parametri del modello stesso, i parametri che descrivono la domanda etc. In un contesto Make-to-Order caratterizzato da una domanda incerta e variabile, può risultare interessante da un punto di vista pratico capire quale è il modello più adatto a seconda delle caratteristiche della domanda. Uno strumento matematico a supporto di questa decisione può essere l'uso di modelli di Machine Learning che possono essere addestrati ad associare il migliore modello a una serie di variabili che descrivono le caratteristiche della domanda. L'obiettivo di questa tesi è testare diverse tecniche di classificazione che hanno l’obiettivo di identificare il modello di rilascio degli ordini che restituisce le migliore performance in termine di Gross Throughput Time medio.
using machine learning techniques as a decision-making tool to support the identification of the best workload control method to implement in a make-to-order context
Maiolli, Samuele
2021/2022
Abstract
Nowadays, manufacturing companies are witnessing a high level of customization required by customers and an increase in competition from low-cost countries. The evolution of the competitive context has prompted companies to compete not only on product price and quality, but also on the service offered to the customer. In fact, companies try to offer customized products and to deliver quickly in compliance with the delivery date negotiated with the customer. In order meet these market demands, companies adopt a Make-to-Order policy, which means that the products are only manufactured from the moment the customer's order arrives. Production planning is therefore a crucial factor for the success of the company. One of the most discussed techniques to deal with this situation is the Workload Control. Workload Control is an order release management tool that aims to keep the workload in the system stable and limited in order to reduce shop congestion and get products flowing quickly through the production process. Workload Control literature is very abundant. In recent decades, many models have been proposed with different characteristics and objectives. However, a dominant model has not been identified since the performance of the models depends on numerous factors such as the configuration of the production system, the parameters of the model itself, the parameters describing the demand, etc. In a Make-to-Order context characterized by uncertain and variable demand, it may be interesting from a practical point of view understanding which model is most suitable according to the characteristics of the demand. A mathematical tool to support this decision can be the use of Machine Learning models that can be trained to associate the best model with a set of variables that describe the characteristics¬ of the demand. The goal of this thesis is to test different classification techniques that identify the order release model that returns the best performance in terms of average Gross Throughput Time.File | Dimensione | Formato | |
---|---|---|---|
2023_5_Maiolli.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
2.52 MB
Formato
Adobe PDF
|
2.52 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/10589/203587