Ship machinery condition monitoring using vibration data through supervised learning

Research output: Chapter in Book/Report/Conference proceedingConference contribution book

393 Downloads (Pure)

Abstract

This paper aims to present an integrated methodology for the monitoring of marine machinery using vibration data. Monitoring of machinery is a crucial aspect of maintenance optimisation that is required for the vessel operation to remain sustainable and profitable. The proposed methodology will train models using pre-classified (healthy/faulty) data and then classify new data points using the models developed. For this, vibration points are first acquired, appropriately processed and stored in a database. Specific features are then extracted from the data and stored. These data are then used to train supervised models pertinent to specific machinery components. Finally, new data are compared against the models developed in order to evaluate their condition. The above will provide a flexible but robust framework for the early detection of emerging machinery faults. This will lead to minimisation of ship downtime and increase of the ship’s operability and income through operational enhancement.
Original languageEnglish
Title of host publicationProceedings of MSO 2016, International Conference on Maritime Safety and Operations
EditorsIraklis Lazakis, Gerasimos Theotokatos
Pages103-110
Number of pages8
Publication statusPublished - 13 Oct 2016
EventInternational Conference of Maritime Safety and Operations 2016 - University of Strathclyde, Glasgow, United Kingdom
Duration: 13 Oct 201614 Oct 2016
http://www.incass.eu/mso-2016/welcome/

Conference

ConferenceInternational Conference of Maritime Safety and Operations 2016
Abbreviated titleMSO 2016
CountryUnited Kingdom
CityGlasgow
Period13/10/1614/10/16
Internet address

Keywords

  • vibration measurements
  • predictive maintenance
  • machine learning
  • condition monitoring
  • SVM

Fingerprint Dive into the research topics of 'Ship machinery condition monitoring using vibration data through supervised learning'. Together they form a unique fingerprint.

Cite this