The automotive industry has been in rapid development which has introduced security  vulnerabilities in the in-vehicle network such as Controller Area Network (CAN) buses. These vulnerabilities expose the vehicle to malicious physical and remote attacks. The CAN bus controls all the communications between the vehicle ECUs,  such as exchanging parameters and the state of the vehicle.  Despite its sensitive role, the CAN bus is the most compromised component in the In-vehicle network as it lacks any form of cryptography methods to ensure integrity and confidentiality of the vehicle’s data. We aim to design a deep learning-based Intrusion Detection System (IDS) compatible with the limitations of the embedded electronics hardware. The proposed model will be trained to detect malicious traffic in the CAN bus indicating security attacks targeting the vehicle, in real time. The IDS will be introduced to different attack classes and tested in real-life environments.

1- Design an intelligent IDS to detect malicious messages on CAN buses.

2- Compress and accelerate the deep learning models used for IDS achieving the suitable performance in line with the limited resources of embedded systems.

3- Deploy the compressed accelerated model on a micro-controller.

4- Test the proposed model in real life environment.

The proposed system is an embedded software developed for intrusion detection for in-vehicle network. The system detects deviant activity in the CAN bus of the in-vehicle network. It is built by deploying an optimized deep learning model on a MCU. The MCU is integrated into the CAN bus of a vehicle as one of its nodes so it can receive and process the CAN traffic for attack detection. This system provides a compressed accelerated model with beneficial processing costs that fits the tight requirements of embedded electronics-based systems.