Embedded Machine Learning-Based Voltage Fingerprinting for Automotive Cybersecurity

This research presents the development and validation of an embedded system for real-time intrusion detection in automotive environments, leveraging voltage-based ECU fingerprinting. The proposed system integrates advanced machine learning algorithms, including SVM, ANN, and DT, optimized for anomaly detection, with linear SVM achieving the highest accuracy. To demonstrate feasibility, the system was implemented on automotive-grade microcontrollers, AURIX Tricore platform TC375. A robust test environment replicating real CAN traffic was employed to validate the approach, based on low-cost embedded devices (Arduino R3/R4 and MCP2515 module). To address synchronization challenges between the CAN transceiver and ADC channels, a circular buffer strategy was introduced, significantly reducing false positives and enhancing system stability. Performance evaluations under nominal and attack scenarios confirm the effectiveness of the solution, offering a practical and reliable method to enhance in-vehicle cybersecurity and mitigate risks for drivers and passengers.