An Integrated TCN-GRU Deep Learning Approach for Fault Detection in Floating Offshore Wind Turbine Drivetrains

In the complex operational environment of offshore wind turbines, the drivetrain system faces multiple uncertainties including wind speed fluctuations, wave disturbances, and dynamic coupling effects, which significantly increase the difficulty of fault identification. To address this challenge, this paper proposes a deep learning model integrating Temporal Convolutional Networks (TCN) and Gated Recurrent Units (GRU) to enhance fault detection capability. The TCN module extracts multi-scale temporal features from vibration signals, while the GRU module captures long-term dependencies in drivetrain degradation patterns. The study utilizes a publicly available Zenodo dataset containing simulated acceleration signals from a 5-MW reference drivetrain under three offshore conditions, covering healthy and faulty states of the main shaft, high-speed shaft, and planet bearings. Experimental validation under different operational conditions demonstrates that the proposed TCN-GRU model outperforms baseline models in terms of accuracy, precision, and recall.