Multi-objective optimization of federated predictive maintenance under energy, bandwidth, and latency constraints in remote wind farms

Abstract

This study introduces a comprehensive multi-objective optimization framework for Federated Learning-based predictive maintenance, specifically engineered for offshore wind energy systems operating under stringent constraints related to energy availability, communication bandwidth, and latency. The proposed framework addresses the complexities of decentralized data environments by enabling edge computing nodes—corresponding to individual wind turbines—to collaboratively develop a global predictive maintenance model while safeguarding data privacy and optimizing resource efficiency. Key architectural components include constraint-aware participation scheduling, adaptive learning rate modulation, and trust-weighted model aggregation, which collectively enhance the model’s predictive fidelity under heterogeneous and intermittent operational conditions. Simulation-based analysis illustrates how the proposed coordination framework can balance competing resource and convergence objectives under constrained participation conditions. The analysis underscores critical trade-offs between accuracy, energy consumption, latency, and communication overhead, highlighting the practical importance of adopting a multi-objective formulation in real-world deployments. These findings contribute to the methodological development of resource-aware federated coordination strategies relevant to remote predictive-maintenance scenarios, providing conceptual insights for future data-driven validation in intelligent asset monitoring systems.