Backup mandate Research Council: Multi-Machine Fault Detection for Wind Turbines using Deep Transfer Learning.

Project Details


With Industry 4.0, wind turbines are instrumented with various wireless sensors to connect with a cloud-based architecture for close monitoring of the machine. Condition monitoring is crucial for wind turbines to anticipate failures to reduce the O&M costs and thus the cost of wind energy. Therefore, early fault detection is essential in reducing operation and maintenance costs and wind turbine downtimes. The evolution of failures is a dynamically-changing and highly complex non-linear process, and thus it is necessary to develop intelligent condition monitoring techniques. We propose a hybrid approach leveraging both knowledge about the machines and the capabilities of AI-driven models. We start from signal processing and fit a fault detection model using deep neural networks. Still, data about fault cases in real-life are scarce. This causes a data imbalance issue for deep learning methods. We aim to solve this problem using multi-machine deep transfer learning, which transforms and exchanges data among similar turbines. Moreover, different signals capture distinct behaviors of wind turbines, and thus may be more sensitive to certain types of faults. We propose to fuse the outcomes of multiple fault models, based on different types of signals (e.g., vibration and temperature), to obtain a reliable hybrid fault detection model. We validate our approach using both lab- generated data, such as the CWRU bearing dataset, as well as real- world wind turbine data from the field.
Effective start/end date1/11/2131/10/22

Flemish discipline codes

  • Machine learning and decision making
  • Knowledge representation and reasoning
  • Neural, evolutionary and fuzzy computation
  • Acoustics, noise and vibration engineering


  • Fault detection
  • Wind Turbines
  • Deep Learning
  • Transfer Learning