A Scalable V2X Architecture for Predictive Maintenance Utilizing Federated Learning in Smart Mobility

Connected vehicles that utilize V2X (Vehicle-to-Everything) communication and have onboard sensing capabilities generate valuable data for predictive maintenance. However, conventional centralized approaches encounter difficulties in terms of data privacy, communication overhead, and scalability. This study presents a scalable vehicle-to-everything (V2X) architecture that utilizes Federated Learning (FL) to facilitate privacy-preserving predictive maintenance. The architecture comprises three tiers: local edge-based vehicle models, a global orchestration platform for model aggregation, and a front-end dashboard for real-time insights. The system was assessed using real test vehicle datasets by comparing FL with centralized learning models. FL achieved comparable or better regression performance (R² = 0.855 vs. 0.822) while maintaining classification accuracy (∆F1 = 0.004). This method maintains the confidentiality of raw vehicle data by only sharing updates related to the model, effectively minimizing the bandwidth requirements. The experiment shows that FL provides a practical alternative to centralized systems in connected vehicle environments, leading to intelligent, decentralized, and secure automotive maintenance solutions. Future work will focus on evaluating the architecture’s performance across different feet sizes to determine scalability.