Abstract This study presents a systematic literature review on the integration of Artificial Intelligence (AI) and Machine Learning (ML) techniques into Prognostics and Health Management (PHM), which integrates condition monitoring, diagnosis, and prognostics to support predictive maintenance decisions. Although several review studies have addressed artificial intelligence-based predictive maintenance, a systematic and structured analysis capable of explicitly identifying research gaps and unexplored combinations within PHM remains limited. Motivated by the increasing relevance of predictive maintenance in Industry 4.0, the review examines how modern AI architectures have been applied to Health State (HS) estimation and Remaining Useful Life (RUL) prediction, which represent, respectively, the current condition of the system and the expected time until functional failure, across critical machinery. To address this gap, the study employs an integrated review framework combining the Systematic Search Flow (SSF), the Morphological Matrix (MM), and the Cross-Consistency Matrix (CCM), which together structure the literature selection, classification of research dimensions, and identification of unexplored variant interactions, enabling a rigorous and multidimensional analysis beyond conventional descriptive reviews. From an initial dataset of 692 publications, 18 primary studies were selected and classified according to four key dimensions: PHM aspect, AI/ML technique, analyzed element, and data-processing interface. The results reveal a strong concentration of research on CNN- and LSTM-based RUL prediction, where Convolutional Neural Networks (CNN) and Long Short-Term Memory networks (LSTM) are used to model spatial and temporal degradation patterns, for bearings, while complex systems, multimodal sensing, embedded intelligence, and IoT-, IIoT-, and AIoT-based interfaces, referring respectively to the Internet of Things, the Industrial Internet of Things, and the integration of artificial intelligence capabilities at the device or edge level, remain largely unexplored. The CCM reveals that only 48% of the 96 possible variant interactions are represented in current literature. This quantitative finding highlights substantial unexplored research opportunities, particularly with respect to real-world deployments, distributed processing architectures, and the development of methodological frameworks for model verification and field validation. Overall, the findings provide a consolidated view of the state of the art and outline a clear research agenda for advancing AI-enabled PHM in industrial environments. By quantitatively mapping these gaps and synthesizing existing approaches, this review contributes a structured perspective that supports the development of practical, scalable, and industrially deployable PHM solutions. As machinery becomes increasingly connected and autonomous, bridging the gap between high-performance predictive models and operational deployment will be essential for realizing the full potential of intelligent maintenance strategies.
Fernandes et al. (Fri,) studied this question.