Analysis of profiles for monitoring of modern ship performance via partial least squares methods

Shipping operators are nowadays facing the challenge of monitoring ship performance based on operational data. This is triggered by the compelling air pollution regulation EU 2015/757 of the European Parliament, which aims from January 2018 to monitoring, reporting, and verification of all harmful emissions of ships operating in the European Economic Area. On the other hand, the continuous acquisition of operational data, which is performed on most of the modern ships, urgently calls for the application of new and opportune statistical methods able to deal with high-dimensional data. Ship operating conditions can be in fact described by sensor signals collected throughout each voyage and stored as profiles. In this paper, the latter are analyzed through multiway partial least squares regression of the average fuel consumption per hour over each voyage, which is chosen as scalar performance response, being proportional to harmful emissions. The proposed approach is able to monitor profiles with different length at different voyages. Nevertheless, it is capable of indicating at which instant anomalies may have occurred in ship operating conditions. The proposed approach is shown to be able to furnish clear indications for supporting prognosis of faults. By means of real data acquired from a Ro-Pax cruise ship owned by the shipping company Grimaldi Group, a different multilinear version that explicitly takes into account the 3-way structure of the data is also compared with the proposed approach.