Evaluation of marine engine load cycles using machine learning methods

Vekios, Dimitrios; Βέκιος, Δημήτριος

dc.contributor.author	Vekios, Dimitrios	en
dc.contributor.author	Βέκιος, Δημήτριος	el
dc.date.accessioned	2020-05-18T12:22:53Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/50604
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.18302
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	Clustering	en
dc.subject	Classification	en
dc.subject	Pre-Processing	en
dc.subject	Load cycle	en
dc.subject	Dynamic time warping	en
dc.subject	Κατηγοριοποίηση	el
dc.subject	Ομαδοποίηση	el
dc.subject	Κύκλος φόρτισης	el
dc.subject	Χρονοσειρά	el
dc.subject	Template	en
dc.subject	Ναυτική μηχανική	el
dc.title	Evaluation of marine engine load cycles using machine learning methods	en
heal.type	bachelorThesis
heal.classification	Marine engineering	en
heal.classification	Machine learning	en
heal.classification	Data analytics	en
heal.dateAvailable	2021-05-17T21:00:00Z
heal.language	en
heal.access	embargo
heal.recordProvider	ntua	el
heal.publicationDate	2019-10
heal.abstract	This diploma thesis focuses on marine engine load cycles analysis using data analytics and machine learning methods. Loading cycles of vessels is a complex field of study that requires consideration of several variables and whose study results can lead to decisions with high economic and environmental impact. The main objective of this thesis is to create a tool that is based on data analytics and machine learning and can properly classify loading cycles into groups based on their similarity and create representative load cycles, which can be used for better understanding of the loading patterns that appear during propulsion. The thesis is divided in two parts. The first part discusses the existing literature on automotive driving cycles, a field that has acquired more attention compared to the study of marine loading cycles. The first part, also, contains the theoretical background and the literature review of data analytics and machine learning concepts, as well as an in-depth analysis of the ones that are used in the practical part, such as data preprocessing, clustering, aggregating and classification algorithms. The theoretical research aims to define the machine learning algorithms that are applied in the second part, which constitutes the main product of the thesis. The second part of the thesis follows four main steps. The first step focuses on the selection/ extraction of the loading cycles’ variables, upon which clustering and classification models will be applied, by calculating the correlation matrix of variables. Pre-processing is, then, applied in the dataset in order to prepare it and make it suitable for algorithmic usage. Furthermore, hierarchical clustering is performed to group the preprocessed timeseries based on similarity criteria. To conclude the number of clusters appropriate for this thesis’ scope, an evaluation measure (average silhouette score) is used that provides an indication of the optimal number of clusters. In the third step, an averaging (or aggregating) algorithm is developed to provide optimal representative loading cycles (or templates) that represent sufficiently the timeseries of each cluster. For this purpose, different distance methods of data analytics are used for templates creation and, then, these methods are assessed based on their results. Finally, the fourth step uses the results of the second step (clustering). More precisely, a machine learning classification algorithm is used to classify the timeseries dataset, which is split into train and test sets. Based on the cross-validation of the outputs of the classifier and the clustering, the accuracy of the classifier is calculated and is, then, assessed. Then, the classifier is modified to achieve most accurate results. In short, the whole model, developed in Python, which is the main product of this thesis, can advise the type (cluster) of a given loading cycle, can divide given loading cycles into groups (clusters) and create representatives of each cluster.	en
heal.advisorName	Papalambrou, George	el
heal.committeeMemberName	Kyrtatos, Nikolaos	en
heal.committeeMemberName	Zarafonitis, George	en
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ναυπηγών Μηχανολόγων Μηχανικών. Τομέας Ναυτικής Μηχανολογίας. Εργαστήριο Ναυτικής Μηχανολογίας	el
heal.academicPublisherID	ntua
heal.numberOfPages	115 σ.
heal.fullTextAvailability	false