Predictive Maintenance (PdM) and Prognostics and Health Management (PHM) are central to Industry 4.0 and Industry 5.0, where intelligent, connected, and human-centered production systems leverage AI, Big Data, IoT, CPS, edge computing, blockchain, and digital twins to enhance reliability, availability, and decision-making. This study presents a combined bibliometric and state-of-the-art analysis of hybrid AI approaches in PdM/PHM, using merged data from Web of Science and Scopus, analyzed with Bibliometrix and VOSviewer, following the PRISMA framework. The analysis emphasizes annual scientific production, source and country contributions, keyword co-occurrence, thematic evolution, and conceptual structure, complemented by a critical review of top-cited documents to extract architectures, methodological trends, and limitations. Several families of hybrid AI architectures are identified, including feature engineering combined with deep learning, CNN-LSTM or CNN-GRU models, autoencoder-based transfer learning, digital twin-assisted deep learning, seq2seq with attention mechanisms, and ensemble learning approaches. The research focuses on predictive maintenance, fault diagnosis, and smart manufacturing, while gaps remain in CPS integration, real-time decision-making, robustness to domain shifts, data scarcity, explainability, and industrial validation, as many studies rely on public datasets. By linking bibliometric insights with technical depth, mapping hybrid AI architectures to PHM functions, and highlighting concrete research gaps such as uncertainty-aware RUL prediction, edge deployment, multi-modal data fusion, and explainable models, this work provides a forward-looking perspective and positions the study as a critical, actionable roadmap for advancing hybrid AI in PdM/PHM within Industry 4.0 contexts.