Propulsion shafting arrangement modeling from mechanical drawings using deep learning and YOLOv8

Ναούμ, Παντελεήμων; Naoum, Panteleimon

dc.contributor.author	Ναούμ, Παντελεήμων	el
dc.contributor.author	Naoum, Panteleimon	en
dc.date.accessioned	2023-06-12T08:47:35Z
dc.date.available	2023-06-12T08:47:35Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/57815
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.25512
dc.rights	Αναφορά Δημιουργού - Μη Εμπορική Χρήση - Παρόμοια Διανομή 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc-sa/3.0/gr/	*
dc.subject	AI	en
dc.subject	Deep Learning	en
dc.subject	Image Processing	en
dc.subject	YOLO	en
dc.subject	Αξονικό Σύστημα	el
dc.subject	ΣΝΔ	el
dc.subject	Αναγνώριση Αντικειμένων	el
dc.subject	Τεχνητή Νοημοσύνη	el
dc.subject	Computer Vision	en
dc.subject	Μοντελοποίηση Αξονικού	el
dc.title	Propulsion shafting arrangement modeling from mechanical drawings using deep learning and YOLOv8	en
dc.title	Μοντελοποίηση αξονικού συστήματος από μηχανολογικό σχέδιο με την χρήση συνελικτικών νευρωνικών δικτύων	el
heal.type	bachelorThesis
heal.generalDescription	In the present thesis, we evaluate the feasibility of developing an Artificial Intelligence algorithm, which will employ the use of Deep Neural Networks, Object Detection and OCR Techniques in order to understand the Propulsion Shaft Arrangement of a vessel, identify and classify key objects in it, identify and extracting dimensions and correlate their positions with their respective dimensions. The purpose of this algorithm is to extract the necessary data to generate a digital model of the vessel’s shafting arrangement capable to undergo further processing such as being incorporated as a module in a Shaft Alignment software. Such an integration will potentially accelerate the Shaft Alignment process instead of manually entering the data into it and further lead a path in scale-up projects such as shaft alignments of multiple vessels in a fleet.	en
heal.classification	Marine Engineering	en
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2023-03-10
heal.abstract	The modern maritime industry is heading towards a digitalized future which encompasses sustainable technological solutions which will automate operations in all aspects of the maritime sector. The role of the Marine Engineer as a professional encompasses the knowledge and the critical thinking required to solve problems and overcome technical difficulties. One of them could be the ability to read mechanical drawings. The Marine Engineer is capable of identifying mechanical objects on a drawing, for example the position of the vessel’s Main Engine on a General Arrangement Plan and matching the identified geometries with their respective dimensions. Thus, the engineer is able to extract the necessary data from a mechanical drawing which are needed to solve further issues and make optimal decisions based on them. In the present thesis, we evaluate the feasibility of developing an Artificial Intelligence algorithm, which will employ the use of Deep Neural Networks, Object Detection and OCR Techniques in order to understand the Propulsion Shaft Arrangement of a vessel, identify and classify key objects in it, identify and extracting dimensions and correlate their positions with their respective dimensions. The purpose of this algorithm is to extract the necessary data to generate a digital model of the vessel’s shafting arrangement capable to undergo further processing such as being incorporated as a module in a Shaft Alignment software. Such an integration will potentially accelerate the Shaft Alignment process instead of manually entering the data into it and further lead a path in scale-up projects such as shaft alignments of multiple vessels in a fleet. Multiple Object Detection algorithms and different OCR models have been proposed throughout computer vision applications such as R-CNN, SSD, RetinaNet, thus we need to employ the necessary metrics to evaluate the most accurate fitted to our application. On this particular thesis, the YOLOv8 (You Only Look Once – Version 8) has been employed, a cutting-edge, state-of-the art model (released at January 2023), which is an improvement of the previous YOLO versions introducing new features and improvements to further enhance performance and accuracy. Additionally, a large database of Shaft Arrangement drawings was generated by performing augmentations, emulating noise, cropping and minimal rotations of the image. We generated the necessary bounding boxes for the classes including several objects necessary to identify such as the Propeller, Flanges, the Stern Tube Bearings, the Shaft Bearing as well as the Propeller Shaft, the Intermediate Shaft and their dimensions. The chosen model was trained for 200 epochs and achieved an average of all classes Precision of 92.4%, Recall of 94.2% and mAp of 94%.	en
heal.advisorName	Papadopoulos, Christos	en
heal.committeeMemberName	Papadopoulos, Christos	en
heal.committeeMemberName	Papalambrou, George	en
heal.committeeMemberName	Anyfantis, Konstantinos	en
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ναυπηγών Μηχανολόγων Μηχανικών. Τομέας Ναυτικής Μηχανολογίας	el
heal.academicPublisherID	ntua
heal.numberOfPages	115 σ.	el
heal.fullTextAvailability	false