Cluster Based Anomaly Detection with Applications in the Maritime Industry
Glad, Ingrid Kristine
Appears in the following Collection
Proceedings of the 2017 International Conference on Sensing, Diagnostics, Prognostics, and Control SDPC 2017. 2017, 328-333,
A modern ship is a highly complex system, often equipped with thousands of sensors to monitor various features of the system. Traditionally, sensor based component control is rule-based. A temperature threshold could for example be predefined, forcing the system to automatically shut down if the temperature surpasses the threshold. The problem with the rule-based approach emerges when we want to analyse multiple signals, and base our decisions on the combined behaviour. The problem space grows rapidly, making it almost impossible to describe a rule base that covers every permutation . Another problem is the lack of fault data. Data from normal operating conditions are continuously collected, while, obviously. data from known faults are more rare.
Due to this, anomaly detection frameworks are designed for detecting unexpected behaviour of a system by comparing it to training data, which represents normal conditions. The unexpected behaviour can represent a fault in the system, but in principle, any behaviour that deviates from the behaviour represented in the training data can be discovered.
To demonstrate the capabilities of the proposed method, we apply the anomaly detection framework for the sensor based surveillance and fault detection on an ocean going ship in operation. The data we have at hand does not contain any known faults, hence we alter the signals to mimic some quite subtle anomalies. The results from the cluster based and the original framework are presented and compared.
The remainder of the paper is structured as follows. The original methodology framework is presented in section II. In section III, the proposed modifications are described. A case study is presented in section IV, where both the original and the modified methodology are demonstrated. Finally, in section V we conclude.
This proceedings paper has been accepted and published in Proceedings of the International Conference on Sensing, Diagnostics, Prognostics, and Control (SDPC). © 2017 IEEE
Search all of DUO
Only this Collection
For students / employees
Submit master thesis
Access to restricted material
All of DUO
Communities & Collections
By Issue Date
By Issue Date
For library staff
View Usage Statistics