A data-driven approach for fault detection of offshore wind turbines using random forests

Compared with onshore wind turbines, fault detection and isolation (FDI) process is more important for offshore ones due to both additional loadings and maintenance difficulties. FDI will be more demanding when it comes to deep-sea floating wind turbines. In this work, an ensemble learning method, random forests (RF), is proposed to perform fault detection of offshore wind turbines, as RF is robust to overfitting, producing not only accurate and quick classification, but also importance ranking for each individual feature. At the same time, supplementary dominant signals are determined for each fault through principal component analysis. The NREL FASTv8 code and OC3-Hywind 5MW floating wind turbine baseline model are used to verify this proposed data-driven FDI design.