Fault detection and degradation trending using hybrid intelligence techniques and multi-source transfer learning

The increasing interest in renewable energy sources directed research to make renewable energy more competitive. Wind energy is expected to provide the largest contributor to renewable energy in Europe, especially in Belgium, which is the fourth-largest offshore wind energy producer in the world. The O&M cost contributes a significant part to wind energy prices. To reduce O&M costs, early fault detection is crucial. The evolution of faults is a dynamically changing and complex non-linear process, and thus it is necessary to develop intelligent condition monitoring techniques. A hybrid fault detection approach is proposed that leverages engineering knowledge and AI capabilities. The AI models are trained on signal processing features to predict progressing faults and to compensate for fault data scarcity, data-inferred knowledge is extracted and shared among multiple sources, such as similar wind turbines and synthetic data. Different signals capture distinct behaviours of wind turbines and may be more sensitive to certain types of faults. Finally, the outcome of multiple signals (e.g. vibration and temperature) fault models are fused to obtain reliable health indicators. The research is validated on real-world offshore wind farm data. It is also validated on open-source lab-generated fault data to compare with state-of-the-art methods. The research outcomes of this project provide offshore wind operators with better plan maintenance strategies, leading to O&M costs reduction.