<![CDATA[Artificial Intelligence-Driven Predictive Maintenance (AI-PdM) has become a strategic approach in smart manufacturing to reduce unscheduled downtime, lower maintenance costs, and extend the lifespan of production equipment. This article presents a conceptual review that synthesizes the latest developments in artificial intelligence algorithms—including classical machine learning, deep learning (CNN, LSTM, CNN-LSTM hybrid), federated learning, and explainable artificial intelligence (XAI) in the context of predictive maintenance in Industrial Internet of Things (IIoT) and digital twin environments. Unlike systematic literature reviews (SLRs) that employ formal selection and reporting protocols, this article uses a narrative synthesis approach and the development of a conceptual framework to map the AI-PdM workflow from sensor data acquisition to maintenance decision support system integration. The novelty of this review lies in the integration of three dimensions rarely discussed simultaneously within a single framework: predictive accuracy, model transparency through XAI, and data sovereignty through federated learning. The synthesis results show that deep learning-based hybrid models consistently outperform conventional methods on failure detection and remaining life estimation tasks, while the integration of XAI and federated learning is becoming an increasingly relevant direction to support the implementation of trustworthy, secure, and compliant data governance in distributed manufacturing environments  ]]>
