Robust fault detection of wind energy conversion systems based on dynamic neural networks

Area of Science:

• Engineering
• Computer Science
• Renewable Energy

Background:

• Faults in wind energy conversion systems (WECSs) are inevitable and can lead to significant economic losses and safety concerns.
• Effective fault detection systems (FDS) are crucial for ensuring reliable operation, preventing damage, and facilitating timely repairs in WECSs.
• Recurrent neural networks (RNNs) are powerful tools for modeling complex dynamical systems and have shown promise in FDS applications.

Purpose of the Study:

• To propose a novel fault detection system (FDS) for wind energy conversion systems (WECSs) utilizing dynamic recurrent neural networks (RNNs).
• To develop an FDS capable of detecting faults in critical components such as generator angular velocity sensors, pitch angle sensors, and pitch actuators.
• To enhance the robustness of the FDS through the implementation of an adaptive threshold mechanism.

Main Methods:

• A comprehensive dynamic model of the WECS, encompassing both mechanical and electrical components, was developed.
• Dynamic recurrent neural networks (RNNs) were employed to create a neural model emulating the normal operational behavior of the WECS.
• The FDS identifies faults by comparing the outputs of the real WECS with those predicted by the dynamic RNN model.

Main Results:

• The proposed FDS successfully detected faults in the generator's angular velocity sensor, pitch angle sensors, and pitch actuators.
• The FDS demonstrated robustness through the use of an adaptive threshold, leading to a low rate of false and missed alarms.
• Simulation results confirmed the capability of the proposed scheme to detect faults promptly.

Conclusions:

• The developed FDS based on dynamic RNNs provides an effective solution for fault detection in WECSs.
• The system's ability to accurately and rapidly detect faults contributes to enhanced safety, reliability, and economic efficiency in wind energy operations.
• The use of an adaptive threshold significantly improves the FDS's reliability by minimizing false and missed alarms.