Artifical neural network based wing optimization

The modern aircraft design is increasingly driven by environmental and operational restrictions. The reason for this is the increasing impact of ecological issues due to global warming. For this reason, regulations and industrial measures were taken to make air traffic greener. These restrictions and environmental targets are performed in terms of aircraft design can be achieved by increasing aerodynamic efficiency. The aim of this study is to maximize aerodynamic efficiency by finding optimal values of sweep angle, taper rate, twist angle and wing incidence angle parameters used in wing design by keeping wing area and span constant. Within the scope of the thesis, a total of 7-parameter optimization study was carried out using 4 pieces twist angle and 1 each of the other design parameters. The time cost of finding optimal wing values using gradient-based and evolutionary algorithm methods, which are traditional optimization methods, is quite high. For this reason, an artificial neural network-based approximation model has been developed. A total of 82 Computational Fluid Dynamics analyses must be performed in order to create a surrogate model using Box - Behnken design, one of the experimental design methods. Computational fluid dynamics (CFD) analyzes required to create the data set were performed by solving Reynolds Averaged Navier Stokes (RANS) equations in ANSYS Fluent. The created data set was properly trained using a feed forward neural network and a surrogate model was created. 2 different optimization studies were carried out using the surrogate model created. In the case where the first of these, the lift coefficient of the optimum wing, can fall below the lift coefficient of the baseline design, an increase of 10,7397% in aerodynamic efficiency is achieved. In the second optimization case, the lift coefficient of the optimum wing is limited to the lift coefficient of the baseline design, and in this case, an increase in aerodynamic efficiency 10,65% is achieved. In the optimum results, the error in aerodynamic efficiency between the artificial neural network and CFD analysis in the 1st and 2nd optimization cases was obtained as 1,5271% and 0,4427%, respectively. This situation shows that the surrogate model can make predictions with acceptable accuracy. Finally, in the 1st and 2nd optimization cases, the fuel consumption of the aircraft will decrease as these increases in aerodynamic efficiencies will provide a 10,739% and 10,6489% reduction in the required thrust, respectively.