ANOBADA (Anomaly Detection on Vehicle Operational Data)

Purpose and goal

The project goal has been to develop statistical methods for clustering and anomaly detection of vehicle condition data. The data is collected during operation of the vehicle, and is accumulated in scalars, vectors, and matrices. The elements in these have strong statistical dependencies which must be handled to achieve correct analysis results. In the project we have investigated and developed a number of methods for clustering and anomaly detection of this kind of data.

Expected results and effects

With the increased amount of collected data from vehicles, there is an increased demand for efficient and scalable methods to analyse and utilize this data. The project has developed methods for clustering and anomaly detection, which will increase the capabilities of the transportation industry to utilize the information in vehicle condition data. The project has also contributed to the research frontier, and to an intensified collaboration between the partners which may lead to future projects.

Planned approach and implementation

The project has used a broad apprach, where several methods has been evaluated to see which are most suitable for clustering and anomaly detection of vahicle condition data. The data that was used in the project comes from 91 vehicles from which condition data had been collected about once a week from July 2013 to February 2016. Demonstration on real data from a new selection of vehicles shows that the developed methods indeed manages to find interesting anomalies of them.