Anomaly Detection for Vehicle Networks

Abstract

A modern vehicle has approximately 100 electronic control units (ECU) that are con- nected through different automotive communication protocols (e.g. CAN, LIN, Flexray, and Ethernet). The connectivity interfaces of the vehicle are growing fast driven by the integration of advanced technology. This increase in vehicle connectivity makes the vehicle network vulnerable to cyber-attacks. However, The current network design, communication protocols (especially CAN protocol), and software practices were never intended to be used in a potential hostile environment.

Today, Cyber-security plays a vital role in our daily life in protecting our smartphones, laptops, and even our vehicles. Vehicle security is a great concern for protecting the life of passengers and pedestrians. However, the vehicle cyber-security does not advance at the same rate as technological systems integration in the vehicle network. This increases the vulnerability and potential attacks against the vehicle. Recently, the necessity to provide solutions to protect the vehicle against cyber-attacks increases significantly.

This research helps to protect the vehicle network against cyber-attacks. It introduces the applied approaches in the automotive AUTOSAR standard for detecting cyber- attacks. Then, it proposes an anomaly detection system for detecting anomalies in the content of the received messages. The implementation uses the Long Short Term Memory (LSTM) neural network for detecting anomalies in malicious messages. The proposed approach relies on training a model to learn the relation and the rate of change of the signal values in the content of the messages. Then our model can detect anomalies based on the learned legitimate behavior of the different messages. A thorough evalua- tion of the proposed model is presented on a generated dataset where different types of data anomalies are introduced.