AI-Driven Predictive Maintenance Frameworks for Software-Defined Electric Vehicles Using Edge Diagnostics

Abstract

The rapid paradigm shift toward Software-Defined Electric Vehicles (SD-EVs) introduces unprecedented complexity in vehicle health management, battery longevity, and real-time powertrain reliability. Traditional cloud-centric diagnostic architectures suffer from excessive latency, high cellular bandwidth consumption, and privacy vulnerabilities when processing high-frequency telemetry streams from distributed in-vehicle sensors. This paper proposes a comprehensive AI-driven predictive maintenance (PdM) framework tailored specifically for SD-EVs, leveraging decentralized edge diagnostics and lightweight machine learning models. By deploying quantized temporal convolutional networks (TCNs) and isolation forests directly onto automotive Electronic Control Units (ECUs) and domain controllers, the proposed architecture achieves real-time anomaly detection and remaining useful life (RUL) estimation for Lithium-ion battery packs, traction inverters, and brushless DC motors. Experimental evaluations utilizing real-world drive cycle datasets demonstrate that our edge-centric diagnostic framework reduces cloud telemetry overhead by 76.4% while maintaining an anomaly detection accuracy of 98.2% and limiting inference latency to under 14 milliseconds per evaluation window. Furthermore, the framework integrates over-the-air (OTA) model updating protocols to ensure continuous adaptive learning without compromising vehicle safety or cybersecurity compliance. This research establishes a robust blueprint for next-generation autonomous and connected electric mobility maintenance.

KEYWORDS: Software-Defined Electric Vehicles, Predictive Maintenance, Edge Diagnostics, Machine Learning, Battery Health Monitoring, In-Vehicle Networks, Over-the-Air Updates.