The security of CAN communication remains an important challenge in automotive cybersecurity, as there is no single standardized approach for securing it. As vehicles become increasingly connected, the number of potential entry points for attackers also increases. The aim of this thesis was to develop a simplified simulation environment for an in-vehicle network, enable the execution of different attack scenarios, and develop a simple, explainable, and computationally lightweight system for detecting anomalies in CAN communication.
A virtual in-vehicle network consisting of multiple electronic control units was developed, modelling the processes and characteristics of CAN communication at the data-link layer that are relevant to the execution and detection of attacks. Network traffic is transmitted in real time over a virtual CAN bus. Five attack scenarios representing different forms of interference with CAN communication were implemented. Based on established approaches, timing-based and entropy-based anomaly detection methods were implemented. Both methods build reference profiles from attack-free traffic and compare them with the characteristics of observed traffic. A graphical user interface was also developed for configuring and running the simulation and controlling the intrusion detection system.
The evaluation showed that both methods operate according to their design within the developed simulation environment and detect different characteristics of attacks, making them complementary in traffic analysis. The developed solution is limited to detecting the effects of attacks on CAN communication at the data-link layer, while the simulation environment represents a simplified model of a real in-vehicle network. Further development could include modelling timing jitter, expanding the network with a larger number of electronic control units, determining detection parameters more systematically, and testing the system on a physical CAN network.
|