The evolution of the automotive industry toward connected software-defined vehicles demands a shift from reactive to proactive, data-driven maintenance strategies. Predictive Maintenance (PdM) addresses this need but faces challenges in terms of data quality, model accuracy, and stakeholder trust. This study presents a next-generation, cloud-native PdM framework designed to overcome these barriers. It introduces an end-to-end reference architecture that integrates secure data ingestion, unified analytics, and scalable ML Ops to manage the entire AI lifecycle. We benchmark classical machine learning models (Random Forest, XGBoost) and deep learning approaches (LSTMand transformer) for key automotive use cases, including high- accuracy fault classification and Remaining Useful Life (RUL)estimation, using industry-standard datasets. To enhance transparency, Explainable AI (XAI) techniques, such as SHAP,are applied to convert complex model outputs into technician- friendly insights. The framework further leverages generative AIfor synthetic data augmentation, automated report generation, and LLM-powered virtual assistants to support diagnostics and decision-making. The novel contributions of this study include the integration of federated learning with differential privacy, component-level digital twins, and an adaptive orchestration layer for over-the-air (OTA) model updates in zonal ECU architectures. This study provides a technical blueprint for building a scalable, intelligent, and privacy-preserving PdM solution that improves vehicle reliability, reduces unplanned downtime, and aligns with modern enterprise strategies in the electric vehicle sector.
Pandey et al. (Mon,) studied this question.