Abstract
This work proposes a method using a long short-term memory neural network as a diagnostic tool to detect wind turbine rotor mass imbalance. The method uses the synthetic minority oversampling technique for data augmentation in an unbalanced dataset. For this purpose, a 1.5 MW three-bladed wind turbine model was simulated at Turbsim, FAST, and Matlab Simulink to generate rotor speed data for different scenarios, simulating different wind speeds and creating a mass imbalance by changing the density of the blades in the software. Features extraction and power spectral density were also used to improve the Neural Network results. The results were compared to nine different classifiers with four different combinations of datasets and demonstrated that the technique is promising for mass imbalance detection.