On the optimal design of fully identifiable next-generation in-vehicle networks

Due to the emerging advances in connected and autonomous vehicles, today's in-vehicle networks, unlike traditional networks, are not only internally connected but externally as well, exposing the vehicle to the outside world and making it more vulnerable to cyber-security threats. Monitoring the in-vehicle network, thus, becomes one of the essential and crucial tasks to be implemented in vehicles. However, the closed- in nature of the vehicle's components hinders the global monitoring of the in-vehicle network, leading to incomplete measurements, which may result in undetected failures. One solution to this is to use network tomography. Nevertheless, applying network tomography in in-vehicle networks is not a trivial task. Mainly because it requires that the in-vehicle network topology should be identifiable. To this end, we propose in this work an identifiable in-vehicle network topology that enables overall monitoring of the network using network tomography. The new topology is proposed based on extensive analysis to ensure full identifiability under the constraint that only edge nodes can monitor the network, which is the case for in-vehicle networks where internal nodes are not directly accessible. We propose two main algorithms to transform existing in- vehicle network topologies. The first algorithm applies to an existing topology which can be transformed into full identifiability by adding extra nodes/links. Evaluation results show the effectiveness of the proposed transformation algorithms with a maximum added weight of only 3% of the original weight. Furthermore, a new optimisation algorithm is also proposed to minimise the topology weight whilst maintaining the full identifiability by redesigning a new topology. With this algorithm, the results show that the total weight can be reduced by 6%. In addition, compared with the existing approaches, monitoring the in-vehicle networks with the proposed approach can achieve better monitoring overhead and a 100% identifiability ratio.