A Light-Weighted Machine Learning Based ECU Identification for Automative CAN Security

EasyChair Preprint 10242

Abstract

The rise of artificial intelligence brings information security challenges for intelligent connected vehicles. Securing the CAN is crucial to ensuring the overall security of the in-vehicle network. Traditional cryptography technology faces challenges of low computational efficiency and excessive data load when identifying ECU. This paper proposes a light-weighted machine learning based identification algorithm that leverages the physical characteristics of ECU. By analyzing the CAN voltage signals in the time and frequency domains, reducing the data load and choosing a suitable classification model, this method achieves high accuracy, high efficiency and low load for safety identification in-vehicle networks. The experimental results on the data sets of both actual vehicles and CAN bus prototypes have verified the rationality and feasibility of the method.

Keyphrases: CAN, ECU identification, light-weighted, machine learning, physical characteristics

@booklet{EasyChair:10242,
  author    = {Jini Li and Man Zhang and Yu Lai},
  title     = {A Light-Weighted Machine Learning Based ECU Identification for Automative CAN Security},
  howpublished = {EasyChair Preprint 10242},
  year      = {EasyChair, 2023}}