Vessel's Performance Modelling

Developing a digital twin for the propulsion system, a Spliethoff group case

More Info


In this thesis, it was examined how high-frequency operational data collected from an automated recording system, can be used to develop a vessel’s performance model. Both the traditional and the Machine Learning (ML) approach were examined. In the traditional approach, every component of the propulsion plant was examined separately, while empirical or semi-empirical methods were used for the added resistance due to waves, wind and shallow water. Historical data were used for verification and tuning (where needed) during the development of the model. In the Machine Learning approach, the method of Support Vector Machines was used. Also, it was examined whether a vessel’s performance model can be developed by combining the methods mentioned above. Thus, in total four models were developed. The data regarding the last year of operation of the vessel Schippersgracht were used (45000 data points) to test the models. It was found that with all the four models, 80 % of the examined dataset was calculated with an error of less than 10 %. The daily fuel consumption was calculated for the same dataset and for 90% of the days examined, the daily fuel consumption was calculated with an error of less than 10 %. The main conclusions about the developed performance models is that all the models resulted in more less the same accuracy. The model based on the traditional approach has the advantage that it offers more information and more outputs can be taken, compared to the ML approach which only provides the output of the fuel consumption. On the other hand, the model based on the ML approach is significantly faster and allows many iterations to be performed in a short time. Regarding the models which were developed by combining the approaches mentioned above, they did not result in additional advantages. Finally, two case studies were performed. In the first case study, it was presented how the model, developed following the traditional approach, can be used to examine how modifications in the propulsion system will influence the vessel’s fuel consumption. The case of switching from fixed rpm (that is currently used) to a variable rpm system was examined, which is a real scenario that the company is considering. It was calculated that installing the variable rpm system and a frequency converter will result in annual savings of around 250,000 € while the investment’s payback time is estimated to be around two years. In the second case study, a future voyage simulation algorithm was developed. This algorithm receives as inputs the voyage plan(route and speed) and performs a calculation for the voyage’s fuel consumption by taking into account the weather predictions. The voyage simulation algorithm is making use of the ML model. A few cases were examined in order to examine how many days in advance the calculation is reliable (due to the uncertainty of the weather predictions) and about the time interval that should be used in the algorithm. However, clear conclusions were not derived, and further examination is suggested(by examining more voyages).