Predictive maintenance leveraging machine learning for time-series forecasting in the maritime industry

• Published in: 2020 IEEE 23rd International Conference on Intelligent Transportation Systems (ITSC) pp. 1 - 8
• Main Authors: Makridis, Georgios, Kyriazis, Dimosthenis, Plitsos, Stathis
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20.09.2020
• Subjects: Anomaly detection; Data models; deep learning; Engines; Fuels; gradient boosting; Machine learning; maritime; Oils; permutation entropy; predictive maintenance; Sensors; time-series
• DOI: 10.1109/ITSC45102.2020.9294450

One of the key challenges in the maritime industry refers to minimizing the time a vessel cannot be utilized, which has multiple effects. The latter is addressed through maintenance approaches that however in many cases are not efficient in terms of cost and downtime. Predictive maintenance provides optimized maintenance scheduling offering extended vessel lifespan, coupled with reduced maintenance costs. As in several industries, including the maritime domain, an increasing amount of data is made available through the deployment and exploitation of data sources, such as on board sensors that provide real-time information. These data provide the required ground for analysis and thus support for various types of data-driven decision making. In the maritime domain, sensors are deployed on vessels to monitor their engines and data analysis tools are needed to assist engineers towards reduced operational risk through predictive maintenance solutions that are put in place. In this paper, we present an approach for anomaly detection on time-series data, utilizing machine learning on the vessels sensor data, in order to predict the condition of specific parts of the vessel's main engine and thus facilitate predictive maintenance. The novel characteristic of the proposed approach refers both to the inclusion of new innovative models to address the case of predictive maintenance in maritime and the combination of those different models, highlighting an improved result in terms of evaluation metrics.