Investigation of reinforcement learning for  optimal part dispatching by drones in flexible  manufacturing systems

Αγγελίδου, Χαρίκλεια; Angelidou, Charikleia

dc.contributor.author	Αγγελίδου, Χαρίκλεια	el
dc.contributor.author	Angelidou, Charikleia	en
dc.date.accessioned	2022-09-27T09:23:15Z
dc.date.available	2022-09-27T09:23:15Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/55769
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.23467
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc-nd/3.0/gr/	*
dc.subject	Reinforcement learning	en
dc.subject	Flexible manufacturing	en
dc.subject	Drones	en
dc.subject	Dispatching	en
dc.subject	Ενισχυτική μάθηση	el
dc.subject	Discrete events	en
dc.subject	Διακριτά γεγονότα	el
dc.subject	Ανάθεση καθηκόντων	el
dc.subject	Ευέλικτη παραγωγή	el
dc.title	Investigation of reinforcement learning for optimal part dispatching by drones in flexible manufacturing systems	en
heal.type	masterThesis
heal.classification	Mechanical Engineering	en
heal.classification	Reinforcement Learning	en
heal.classification	Manufacturing Engineering	en
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2022-04-07
heal.abstract	The industrial environment of the past years has been characterized by the high rate of changes, pushing the industry to implement innovative technologies to satisfy the market needs. It is of high priority to modify existing production systems in order to adopt to the new era and optimize product control. Industry 4.0, the conjunction of Information Technology with well-established means of mass production, is gaining ground each passing day. Detailed simulations simplify the analysis processes and thus allow for a better understanding of the systems, while at the same time decreasing costs and leading to broad optimization possibilities. Unmanned aerial vehicles (UAVs) have also witnessed increasing use, as they provide flexibility and agility to the manufacturing processes, meeting the enhanced efficiency demands. The environmental impact and economic restraints pave the way towards on demand manufacturing, so that the transportation scheme is also transforming to an efficient dispatching network, where dynamic and stochastic environments are required. Quick response to customer orders and the ability to work over a disparate set of floor settings is of high priority. In this study, we address this problem of dispatching in manufacturing. A design to formulate the shop floor state as a virtual discrete events model using Reinforcement Learning (RL) is proposed. Machine processing times, transition times, queue warehouse condition and drone battery are incorporated in the RL state representation, while a reward function is defined in terms of production maximization. The influence of different model parameters and hyper parameter optimization on the quality and stability of the results obtained is analyzed. A well configured Proximal Policy Optimization (PPO) algorithm, that enables optimal part dispatching within the Flexible Manufacturing System is also discussed. The software package MATLAB SimEVents is used for modeling the manufacturing cell environment, interfaced with the MATLAB RL Toolbox for agent creation and training. The results obtained suggest a robust approach to optimizing part dispatching in production lines	en
heal.advisorName	Βοσνιάκος, Γεώργιος - Χριστόφορος	el
heal.advisorName	Vosniakos, George - Christopher	en
heal.committeeMemberName	Βοσνιάκος, Γεώργιος - Χριστόφορος	el
heal.committeeMemberName	Μπενάρδος, Πανώριος	el
heal.committeeMemberName	Τόλης, Αθανάσιος	el
heal.committeeMemberName	Vosniakos, George - Christopher	en
heal.committeeMemberName	Benardos, Panorios	en
heal.committeeMemberName	Tolis, Athanasios	en
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Μηχανολόγων Μηχανικών	el
heal.academicPublisherID	ntua
heal.numberOfPages	99 σ.	el
heal.fullTextAvailability	false