In-Vehicle Network Monitoring with Network Tomography

Abstract

With the advances in Connected and Autonomous Vehicles (CAVs), the in-vehicle network becomes more complex and harder to manage. In addition, as the vehicle becomes part of Internet-of-Vehicles (IoVs), it is now more exposed to the outside world than ever. Current in-vehicle networks are vulnerable to cyber-security threats, including IP-based attacks. Monitoring the in-vehicle network is thus one of the crucial tasks that deserves careful consideration. However, the closed-in system of the in-vehicle network entails difficulty in accessing the internal elements of the network. This hinders the monitoring process from obtaining insights about the internal performance of the network. As a result, monitoring every part of the network becomes intractable. Our focus in this thesis is to investigate monitoring solutions that do not require contribution from the internal components of the network. To this end, our first contribution is that we propose, for the first time in literature, to use network tomography as a monitoring approach for in-vehicle networks. An attractive feature of network tomography is that it exploits the end-to-end measurements to infer the performance of an internal network without the need to access any internal component. This feature is well-suited for closed-in systems such as in-vehicle networks. One challenge in applying network tomography in an in-vehicle network, however, is that the in-vehicle network must be fully identifiable. The second contribution of this thesis is to address this challenge by leveraging the advances in deep learning and proposing to complement network tomography with deep learning-based tomography. Furthermore, to facilitate the monitoring and management process, the third contribution is to propose a new in-vehicle network topology that is fully identifiable, redundant and enabled with Software-Defined Networking (SDN) functionalities. Such topology is aligned with the next-generation Electronic and Electrical (E/E) architecture that is shifting towards centralisation. The proposed monitoring approach can be applied to such topology to infer the overall performance of the network. In addition, with the SDN paradigm and redundancy support, the network can intelligently cope with failures by locating the failed component and re-routing the traffic to the redundant alternative. Hence, achieving self-healing in-vehicle network without external intervention.