Anomaly Detection System for Altered Signal Values within the Intra-Vehicle Network

A modern vehicle is a complex system of sensors, actuators, and electronic control units (ECUs) connected through different automotive networks. In the past, vehicle networks used to be isolated from the outside world which made them immune to attacks. However, recent technologies have made vehicles vulnerable to cyberattacks. This paper proposes a supervised prediction model that allows the ECUs to detect anomalies in the content of the received messages. The implementation introduces a novel approach to rely on the signal value instead of the bitstream values for feature selection. The advantage of this approach will be highlighted. Additionally, a discussion of anomaly detection in the AUTOSAR standard is presented showing how the implemented model can be integrated into a network of ECUs running AUTOSAR communication stacks. A thorough evaluation of the proposed model is presented on a generated dataset where different types of data anomalies are added.