Universal Intrusion Detection System on In-Vehicle Network

The Controller Area Network (CAN) protocol is widely used in automotive and industrial applications for communication. However, the lack of authentication and encryption in CAN bus networks has made them vulnerable to cyberattacks. This study investigated the effectiveness of different intrusion detection models in accurately classifying attacks, such as Denial-of-Service (DoS) attacks, fuzzing, and replay attacks. A labeled dataset was created using a methodology that uses the CAN ID sequence, time gap, and hamming distance between hexadecimal strings of equal length. The resulting dataset was segmented and converted into heat maps that were input to deep learning models such as VGG-16, AlexNet, and ResNet-50. The study provides valuable insights for developing more robust security measures for in-vehicle networks. However, recent research has shown that intrusion detection systems need to be developed individually for each vehicle, taking into account the unique data characteristics of the vehicle. Therefore, this paper proposes to implement universal IDS by using three types of CNN architectures to find the best one that is suitable for all types of attacks with high accuracy.