Enabling Predictive Maintenance in Smart Buildings: A Review of AI Approaches, Digital Integration, and System Challenges

The shift toward sustainable and intelligent buildings has accelerated the demand for advanced maintenance strategies that reduce energy waste, downtime, and improve reliability. Traditional maintenance approaches (reactive and preventive) are increasingly inadequate for managing the complexity and dynamic behavior of modern building energy systems. Predictive maintenance (PdM) offers a transformative paradigm for real-time monitoring, fault anticipation, and optimized planning. However, its implementation is challenged by fragmented data, limited interoperability, high costs, and insufficient real-world validation. Although PdM has been reviewed, integration of modeling and deployment remains limited. Based on 238 peer-reviewed studies, this paper reviews PdM systems in building energy applications, addressing both the algorithmic and technical aspects of deployment, with a focus on data-driven methods for fault detection, diagnosis, and prognosis. It first examines the challenges within building energy systems that necessitate PdM innovation and then outlines a structured roadmap for developing real-time PdM systems. The review underscores the need for unified frameworks that combine detection, diagnosis, and prognosis with real-time analytics to support smart, resilient building systems. It also surveys trends in real-time monitoring systems and integration strategies to enhance maintenance performance including AI techniques such as transfer learning, active learning, explainable AI, interoperability methods, and large language models, alongside enabling technologies such as digital twins, edge computing and cloud infrastructure. Finally, it examines persistent challenges in data quality, modeling, deployment, operations, and risk, proposing future research directions. The goal is to guide researchers by outlining a step-by-step pathway for advancing PdM in intelligent buildings.