An application of sensor-based anomaly detection in the maritime industry

In this paper we present an application of sensor-based anomaly detection in maritime transport. The study is based on real sensor data streamed from a ship to shore, where the data is analysed through a big data analytics platform. The novelty of this work originates in the use of data from sensors covering different aspects of the ship operation, exemplified here by propulsion power, speed over ground and ship motion in four different degrees of freedom. The developed method employs Auto Associative Kernel Regression (AAKR) for signal reconstruction, and the Sequential Probability Ratio Test (SPRT) technique for anomaly detection, where different hypothesis tests looking both at mean and variance deviations have been tested. In order to compare different settings, formal state of the art performance metrics have been used. We demonstrate that the AAKR model produces good reconstructions when the observations are similar to observations represented in the training data, and for some examples of simulated anomalies, the method reveals the abnormal behaviour. As long as the parameters are tuned carefully, alarms are triggered appropriately by the SPRT.