Parametric studies of aerodynamic properties of wings using various forms of machine learning

In aerodynamic studies, models are essential tools for understanding complex fluid flow phenomena. However, their use can be expensive in terms of computer power and calculation time. Therefore, machine learning algorithms have become essential when it comes to analyzing uncertainty in modelling and predicting the values for new input parameters with sensitivity quantifications and in a reasonably short time. The aim of this paper is to predict the key factors in aircraft design by finding the best estimation of the dependent variable in the form of the lift to drag ratio, for any new input-dependent values in the form of Mach numbers and angle of attack. Therefore, different regressions of classical supervised learning algorithms have been applied. The statistical errors have been calculated for these regressions in order to choose the best fit for an unknown model. In addition, artificial neural networks (ANN) have been used to train the data, and to predict the ratio of lift to drag in a practical time compared to the use of experimental tests and the computational fluid dynamics (CFD) technique.