Anomaly detection in cyber-physical applications
Gutierrez Rojas, Daniel (2023-05-25)
Väitöskirja
Gutierrez Rojas, Daniel
25.05.2023
Lappeenranta-Lahti University of Technology LUT
Acta Universitatis Lappeenrantaensis
School of Energy Systems
School of Energy Systems, Sähkötekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:ISBN:978-952-335-934-5
https://urn.fi/URN:ISBN:978-952-335-934-5
Tiivistelmä
Anomaly detection is a crucial task in many industrial environments as it enables a safer and more reliable operation of different processes; it is usually seen as a separate process from the industrial system design. Performance improvements mostly rely on detection methods. Therefore, anomaly detection can be seen as an end in itself. In this context, the focus of this doctoral dissertation is to highlight the importance of anomaly detection considering different, often neglected, subfunctions such as sampling, network architecture, and planning support. These subfunctions, along with an analysis of detection performance and upcoming communication technologies, are merged to ultimately improve anomaly detection. This is achieved by the integration of anomaly detection in the cyberphysical system composed of three articulated layers: a physical layer, a data layer, and a decision layer, where the layers have specific tasks that could be enhanced or modified, thereby providing additional flexibility and scalability. Simple adjustments to the physical layer like an increase in the number of sensors or changing the sampling methodology may result in an average improvement of the data transmission rate. In the data layer, the data aggregation method to be used has an impact on the anomaly detection accuracy. The results presented in this doctoral dissertation were obtained in specific cyber-physical applications, such as a microgrid, a chemical process, and a power transmission system. The three-layer model applied to the Tennessee Eastman process shows how it benefits the levels of data processing, pointing out in which layer greater improvements can be made even including details of the communication network and the computing platform. In the physical layer, the event-driven method used to transmit samples from the sensors yielded gains in the data transmission rate of 20% and improved anomaly detection in five out of six hard-to-detect faults. In a transmission line application, in the data layer, the quantitative association rule mining algorithm was able to maintain a 98% accuracy of anomaly detection while retrieving explainable results. Furthermore, in the decision layer, predictions of anomalies served as multiobjective chance constraint optimization, balancing resilience and economic objectives in a microgrid. Finally, an extensive analysis of protection in microgrids for anomaly detection showed that multiconnectivity of wired and wireless technologies, such as 5G, meets the requirements of wired networks, thereby improving the flexibility of this application.
Kokoelmat
- Väitöskirjat [1037]