Obtaining Rotational Stiffness of Wind Turbine Foundation from Acceleration and Wind Speed SCADA Data

Area of Science:

• Engineering
• Artificial Intelligence
• Renewable Energy

Background:

• Wind turbine foundation health is critical for operational safety.
• Current methods for assessing foundation stiffness are costly, often requiring specialized sensors.

Purpose of the Study:

• To develop a cost-effective method for estimating wind turbine foundation rotational stiffness.
• To utilize readily available SCADA data (acceleration and wind speed) for foundation health monitoring.
• To enable early detection of foundation deterioration.

Main Methods:

• A convolutional neural network (CNN) model was employed to map acceleration and wind speed data to moment and tilt values.
• Rotational stiffness was estimated by linear regression in the moment-tilt plane.
• An autoencoder generative AI network was used to create synthetic data for re-training the model.

Main Results:

• The CNN model achieved stiffness estimations within 7% of ground truth values on average.
• Re-training with synthetic data allowed for the detection of stiffness drops, though not their exact magnitude.
• The approach offers a significant cost reduction compared to traditional sensor-based methods.

Conclusions:

• The proposed method provides a viable and economical approach to monitor wind turbine foundation stiffness using SCADA data.
• This technique can serve as an early warning system for foundation degradation, improving overall wind farm safety and maintenance.
• Integration of AI in SCADA data analysis offers new possibilities for predictive maintenance in the renewable energy sector.