M.R Soltani - Aerospace Department, Sharif University of Technology, Professor
F Rasi Marzabadi - PhD. Student
M Masdari - PhD. student

Neural networks were used to minimize the amount of data required to predict the aerodynamic coefficients of an airfoil oscillating in plunging motion. For this purpose, series of experimental tests have been conducted on a section of a 660kw wind turbine blade. Two MLP (multi layer perceptron) and GRNN (general regression neural network) were trained using experimental data of the airfoil at various conditions. Results showed that with using only 50% of the acquired data, the trained neural networks were able to predict accurate results with minimal errors when compared with the corresponding measured values. Moreover, these methods can predict the aerodynamic coefficients of the plunging airfoil at different oscillation frequencies, amplitudes, and incidence angles. Therefore with employing this trained networks, the aerodynamic coefficients are predicted accurately with minimum experimental data; hence reducing the cost of tests while achieving acceptable accuracy.

Neural Network-MLP-GRNN-Plunging-wind turbine