Ship diesel engine performance modelling with combined physical and machine learning approach

Ship diesel engine performance modelling with combined physical and machine learning approach

Coraddu, A. (University of Strathclyde)
Kalikatzarakis, M. (Damen Schelde Naval Shipbuilding)
Oneto, L. (University of Genoa)
Meijn, G. J. (University of Strathclyde)
Godjevac, M. (TU Delft Ship Design, Production and Operations)
Geertsma, R.D. (TU Delft Ship Design, Production and Operations; Netherlands Defence Academy)

2018

Condition Based Maintenance on diesel engines can help to reduce maintenance load and better plan maintenance activities in order to support ships with reduced or no crew. Diesel engine performance models are required to predict engine performance parameters in order to identify emerging failures early on and to establish trends in performance reduction. In this paper, a novel approach is proposed to accurately predict engine temperatures during operational dynamic manoeuvring. In this hybrid modelling approach, the authors combine the mechanistic knowledge from physical diesel engine models with the statistic knowledge from engine measurements on a sound engine. This simulation study, using data collected from a Holland class patrol vessel, demonstrates that existing models cannot accurately predict measured temperatures during dynamic manoeuvring, and that the hybrid modelling approach outperforms a purely data driven approach by reducing the prediction error during a typical day of operation from 10% to 2%.

Condition Based Maintenance
Data-Drive methods
Exhaust gas temperature prediction
Gray-Box Models
Machine Learning

http://resolver.tudelft.nl/uuid:c990afcc-20ff-448a-9d40-f8850fb3c7c8

https://doi.org/10.24868/issn.2631-8741.2018.011

Proceedings of the International Ship Control Systems Symposium, 1

14th International Naval Engineering Conference and Exhibition incorporating the International Ship Control Systems Symposium, INEC/iSCSS 2018, 2018-10-02 → 2018-10-04, Glasgow, United Kingdom

Institutional Repository

journal article

© 2018 A. Coraddu, M. Kalikatzarakis, L. Oneto, G. J. Meijn, M. Godjevac, R.D. Geertsma