Abstract
Analyzing the performance of an airfoil or wing is a crucial step in the aircraft design process. Collecting aerodynamic data from numerical or experimental perspectives is time-consuming and costly; however, these challenges can be mitigated through Machine Learning (ML). This study introduces two well-known ML methods for predicting an airfoil's aerodynamic coefficients. Feed-Forward Back Propagation Neural Network (FFBPNN) and Multiple Linear Regression (MLR) are developed using RANS-based numerical simulation data of an airfoil under various flow conditions. The FFBPNN model with 10 hidden layer neurons obtained the lowest RMSE of 0.0082988. The regression lines showed an overall correlation coefficient (R) close to 1. The MLR algorithm's bagging tree regression model produced the best RMSE values of 0.23999 for the lift coefficient and 0.030114 for the drag coefficient. The regression lines had correlation coefficients (R) of 0.85 and 0.84, respectively. Overall, the FFBPNN model demonstrated better results than the MLR method in predicting aerodynamic coefficients, showing higher accuracy and lower prediction errors. This study enhances knowledge of how ML approaches can improve predictions of airfoil coefficients and has practical implications for better aerodynamic designs in the aviation industry.
NETL
NSTL
Springer Nature