Diagnosis Algorithm for Wind Turbine Blade Imbalance based on Nacelle Acceleration

초록

Blade mass and aerodynamic imbalances result in increased fatigue loads in wind turbines. A diagnosis algorithm is proposed to classify these imbalances using nacelle acceleration signals across the full wind turbine operational range. The NREL 5 MW reference turbine is utilized for aeroelastic simulations under turbulent wind conditions ranging from 4 to 20 m/s. Data are generated for seven conditions: one normal case, three mass imbalance cases, and three aerodynamic imbalance cases. The 1P amplitudes are extracted from fore-aft and side-to-side nacelle accelerations using the rotor azimuth angle. These amplitudes, combined with mean blade pitch angle and mean rotor speed, are selected as input features for a supervised neural network. Model hyperparameters are determined via Bayesian optimization. Validation and test accuracies of 96.6% and 97.6% are obtained, respectively. Mass and aerodynamic imbalances are distinguished by the model, and misclassifications are observed primarily between adjacent severity levels within the same fault category. The methodology is verified for the diagnosis of fault type and severity using 5-minute data intervals.

키워드