Till innehåll på sidan
Till KTH:s startsida Till KTH:s startsida

Anomaly detection using first order sliding mode observers

Tid: Ti 2022-11-22 kl 14.00 - 15.00

Plats: Malvinas väg 10, floor 7, Harry Nyquist

Videolänk: https://kth-se.zoom.us/j/63214114459

Språk: English

Medverkande: Twan Keijzer, Delft Center for Systems and Control, TU Delft, The Netherlands

Exportera till kalender

Abstract: Detection of anomalies in cyber-physical systems (CPSs) allows for automated accommodation of the original anomaly, i.e. by means of repairs or reconfigurations in the cyber or physical system. Sliding mode observers (SMOs) have been proposed for exact anomaly estimation for a class of ideal systems without unmatched uncertainties and measurement noise. For such ideal systems anomaly detection is trivial, however for systems with unmatched uncertainties or measurement noise a dedicated detector is needed. In this presentation I will present the design procedure for two of such robust anomaly detectors. These detectors extend the anomaly detection capability of a large class of SMOs to include systems with unmatched uncertainties and measurement noise. The first detector is based on the so-called equivalent output injection (EOI), which is closely related to the anomaly estimate. The second detector is directly based on the SMO state estimation error. Doing so, the second detector bypasses the low-pass filter generating the EOI, allowing for faster detection of anomalies and making it possible to detect smaller magnitude anomalies. The obtained theoretical results are illustrated by application of the detectors to (1) detect a man-in-the-middle (MITM) attack on a collaborative vehicle platoon (CVP) and to (2) detect an oscillatory failure case (OFC) in the servo loop control of a commercial aircraft.

Bio: Twan Keijzer received the MSc degree in Aerospace Engineering from Technical University Delft, The Netherlands, in 2018. He is currently a PhD candidate at Delft Center for Systems and Control under the supervision of Riccardo Ferrari, together with whom he won an Airbus Award at IFAC 2020 for the best contribution to the competition on Aerospace Industrial Fault Detection. His research interests include fault and cyber attack detection methods for aerospace and automotive applications, as well as distributed fault tolerant control approaches