Predictive maintenance of vehicle fleets through hybrid deep learning-based ensemble methods for industrial IoT datasets.

The article focuses on a hybrid deep learning-based ensemble method (HDLEM) developed for predictive maintenance of connected vehicle fleets within Industry 4.0 frameworks. HDLEM integrates three deep learning models—modified Cox proportional hazard deep learning (MCoxPHDL), modified deep learning embedded semi-supervised learning (MDLeSSL), and merged long short-term memory (MLSTM) networks—to predict vehicle time between failures (TBF) using multi-source data, including sensor and historical maintenance records. Experimental results across multiple datasets demonstrate that HDLEM outperforms individual models and various ensemble learning techniques (boosting, bagging, stacking, augmenting) in accuracy, F1 score, sensitivity, and specificity, supporting more efficient, cost-effective, and sustainable fleet management. The study also addresses computational complexity and scalability considerations, highlighting HDLEM's potential for real-world deployment in vehicle fleet predictive maintenance.