Wind Turbine Rotor aerodynamic imbalance detection using CNN

Abstract

This paper investigates the use of convolutional neural networks (CNN) to automatically detect aerodynamic imbalances in horizontal axis wind turbines (WTs). The database is assembled with low-frequency acquisition rates, similar to those obtained using supervisory control and data acquisition (SCADA) systems. The methodology considers imbalances caused by pitch errors only, which might occur due to installation faults or pitch control errors. The measured raw data is initially processed using traditional statistical techniques. Next, the Gramian Angular Field (GAF) method is used to transform the statistical data into images, and then, a CNN is trained to identify aerodynamic rotor imbalance. The proposed methodology is evaluated under numerical simulations of a 1.5 MW wind turbine, and the accuracy and feasibility of the method are demonstrated. The paper demonstrates that it is possible to detect an aerodynamic imbalance in wind turbine rotors from statistics descriptors of nacelle IMU translational accelerations and wind speeds, considering a sampling frequency of above 0.05 Hz, and using an artificial intelligence technique.