AI-based cybersecurity for CAN and IP communication in existing vehicle environments

Purpose and goal:

The purpose of the project was to demonstrate the feasibility of developing a security system that can use machine learning to analyze CAN bus and IP communication for vehicle networks in order to detect abnormal communication behavior. Abnormal behavior can be caused by cyberattacks, such as so-called supply chain attacks where malicious code is already present in the component at delivery, or simply wear and tear or component failures from the factory.

Expected results and effects:

The project has demonstrated that machine learning can be used to develop a security system for the automotive environment for anomaly detection in internal IP and CAN communication networks, without the need for external computing resources. Machine learning can automatically train behavioral models, which simplify integration compared to alternative rule-based methods that require deep insight into expected communication behavior to be configured correctly.

Approach and implementation:

The project followed the plan for the most part, but access to the test bench and physical vehicle was limited. A virtual testing environment based on recorded communication had a positive impact on the project´s results by allowing continuous testing and validation during the project, a necessity for integrating machine learning into the concept. A detailed testing and validation plan was developed to ensure maximum effect given the limited access to the test bench and physical vehicle.