Wind-Turbine Blade Load Prediction & Surrogate

Background

Blade fatigue load determines lifetime and maintenance cycle. Load prediction under extreme winds (turbulence, gusts, shear) is a design challenge; traditional FAST / OpenFAST takes minutes per run, unable to support real-time O&M.

Approach

Simulation data

• Batch OpenFAST runs (various wind conditions, speeds, directions)
• Output: root loads (flap/edge/torsion), power curve

Surrogate

• Random Forest + XGBoost: tabular, interpretable
• 1D-CNN: time-series load signals
• LSTM + Attention: long-range dependence

Key results

• 3-12 m/s: Random Forest < 5%, LSTM < 3%
• 12-25 m/s: Random Forest < 8%, LSTM < 4%
• Extreme turbulence (IEC Ti>25%): Random Forest < 15%, LSTM < 8%

Use cases

• Smart O&M: real-time load evaluation for maintenance windows
• Control optimization: predicted loads for pitch/yaw tuning
• Design: quick comparison of load differences across turbines

Resources

• IEA Wind Task 36 open dataset (multiple turbines)
• NREL 5 MW reference turbine (with OpenFAST)