A Knowledge Graph-Driven Approach for Predictive Maintenance and Fault Diagnosis in Offshore Wind Turbines

Offshore wind turbines (OWTs) represent critical assets in the global transition to renewable energy, yet their operation and maintenance (O&M) costs can constitute up to 30% of the levelized cost of energy due to harsh marine environments, limited accessibility, and complex failure modes in gearboxes, blades, generators, and drivetrains. This review examines a knowledge graph (KG)-driven paradigm for predictive maintenance (PdM) and fault diagnosis (FD) as an exemplar of artificial intelligence (AI) empowerment within intelligent engineering systems (IES). Grounded in cyber-physical systems (CPS) theory, systems engineering principles, and information-theoretic perspectives on uncertainty reduction, the approach integrates heterogeneous data sources—including SCADA streams, vibration signatures, maintenance logs, weather data, and OEM documentation—into semantically structured graphs. These graphs enable causal reasoning, multi-label fault classification, and maintenance action prediction through graph neural networks (GNNs) such as time-domain relational graph convolutional networks (TRGCNs). Concrete implementations, including simulation-augmented KGs achieving over 94% fault-mode accuracy and 90–93% maintenance-action prediction precision for gearboxes, demonstrate substantial reductions in unplanned downtime and vessel deployment costs. The framework exemplifies broader IES transformations: from reactive to proactive strategies, isolated monitoring to integrated CPS ecosystems, and manual interventions to semi-autonomous control augmented by digital twins (DTs) and edge intelligence. Challenges include data heterogeneity, ontology incompleteness in rare-event offshore scenarios, cybersecurity vulnerabilities, and the tension between model expressiveness and real-time latency. Prospects encompass neuro-symbolic hybrids with foundation models, physics-informed embeddings, and multi-agent cognitive CPS architectures. By bridging data-driven pattern recognition with explicit domain knowledge, KG-driven methods enhance explainability, resilience, and lifecycle value in distributed energy systems while highlighting the necessity of standardized data pipelines and human-AI collaboration protocols.