Event-driven Relationship Mining for Operational Technology Monitoring in Intelligent Transportation Systems
Year: 2021
Type: dissertation
Abstract: Critical infrastructures, such as those found in intelligent transportation systems, require extensive monitoring to ensure the correct functionality and overall reliability of their underlying technologies. Commonly, the detection of system anomalies and failures is done by experts who manually examine the event logs. However, this form of monitoring becomes more difficult as systems grow in size and complexity. They produce huge amounts of data that may be spread across many different log files, making it extremely challenging to spot meaningful correlations and failure dependencies by hand. With the recent advances in data analytics, there is a trend towards the automatic analysis of log files to exploit the full potential of the available data, aid in an accurate assessment of the current situation and simplify maintenance. \nThis thesis focuses on technical monitoring by presenting an approach that enables the automatic detection of dependencies between devices and systems solely using the event logs that are already generated and collected during normal operation. Our contribution is threefold: First, a systematic overview of relationship mining approaches and their properties is given and recent literature is compared in order to gain an insight into the current state of the art. Secondly, the relationships between devices are inferred based on temporal coincidence by applying a statistics-based approach to calculate the functional connectivity. This method is extended by calculating correlations on event type level instead of calculating them directly between the devices. Afterwards, the results are aggregated to obtain the desired relationships between devices, which results in an overall increase in precision across all simulated scenarios, in some cases even significantly. Furthermore, enhancing the inferred dependencies by incorporating additional context information, which is supplied in the form of a conceptual model specified by domain experts, is discussed and prototypically implemented. Finally, the third part concludes this thesis by an extensive evaluation of our approach, addressing open challenges and giving an outlook on possible further research directions.
Author: Spitzl, Lisa
Institution: Johannes Kepler University of Linz
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7000508545
