Wind turbine anomaly detection based on SCADA: A deep autoencoder enhanced by fault instances

Area of Science:

• Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Deep autoencoder algorithms are increasingly used for wind turbine condition monitoring and anomaly detection.
• Existing methods often focus on normal data, limiting performance and robustness due to scarce fault instances.

Purpose of the Study:

• To develop an enhanced deep autoencoder that utilizes fault instances for improved anomaly detection in wind turbines.
• To address the challenge of limited fault data by incorporating data augmentation techniques.

Main Methods:

• Developed a triplet-convolutional deep autoencoder (triplet-Conv DAE) integrating convolutional autoencoder and deep metric learning.
• Employed an improved generative adversarial network (GAN)-based data augmentation method to generate synthetic fault instances.

Main Results:

• The proposed triplet-Conv DAE effectively captures normal operation patterns and learns discriminative features using fault instances.
• The method demonstrated superior performance compared to three state-of-the-art anomaly detection techniques.
• The GAN-based augmentation significantly improved triplet-Conv DAE performance with insufficient fault data.

Conclusions:

• The developed triplet-Conv DAE offers a robust and effective solution for wind turbine anomaly detection.
• Integrating fault instances and data augmentation enhances the precision and reliability of condition monitoring systems.