Offshore Wind Turbine Gearbox Fault Identification Using Multi-Domain Feature Fusion and Hybrid Ensemble Learning

Abstract

Gearbox faults are a major cause of downtime and maintenance costs in offshore wind turbines. This paper presents a gearbox fault identification method that combines multi-domain feature fusion with a hybrid ensemble learning framework. Time-domain statistical features and frequency-domain spectral features are extracted and fused to capture both amplitude and frequency characteristics of vibration signals. A hybrid ensemble classifier integrating Support Vector Machine (SVM) and XGBoost is employed to improve classification robustness and accuracy under variable operating conditions. The method is validated using the NREL wind turbine gearbox benchmark dataset. Experimental results demonstrate high fault detection accuracy, outperforming single-domain and single-model approaches.