Data-Driven Detection of Simultaneous Cyberphysical Faults in Connected Electric Vehicle Platoons

Connected autonomous electric vehicles promise improved fuel efficiency and traffic flow through informed decision-making. However, vehicular networks present a wider cyberattack surface for malicious agents and increase the likelihood of simultaneous physical faults. Existing methods for health monitoring and fault diagnosis are inadequate to address this new environment. We propose a two-stage machine learning-based method to diagnose cyberphysical faults—our architecture consists of a lightweight binary classifier to perform the preliminary task of localization, and a multi-head attention module to perform classification. To generalize to arbitrary platoon sizes, we leverage the platoon topology and consider measurements from vehicle pairs. We develop a simulation framework utilizing low-level EV dynamics for the simulation and study of simultaneous faults. On a dataset generated using this framework, we show that our method achieves good performance compared with baseline methods, and we additionally demonstrate robustness to measurement noise.