Airfoil shape optimization using genetic algorithm coupled deep neural networks

To alleviate the computational burden associated with fluid dynamics (CFD) simulation stage and improve aerodynamic optimization efficiency, this work develops an innovative procedure for airfoil shape optimization, which is implemented through coupling genetic algorithm (GA) optimizer coefficients prediction network (ACPN) model. The ACPN established using a fully connected neural geometry as input output. results show that ACPN's mean accuracy lift drag coefficient high up to about 99.02%. Moreover, time of each within 5 ms, four orders magnitude faster compared CFD solver (3 min). Taking advantage fast accurate prediction, proposed model replaces expensive simulations couples GA force change maximize lift–drag ratio under multiple constraints. In terms optimized airfoils can be obtained 25 s. Even considering extra 50 h spent on data preparing 20 s training, overall calculation cost reduced by remarkable 62.1% GA-CFD method (5.5 days). Furthermore, GA-ACPN improves without constraint 51.4% 55.4% NACA0012 airfoil, respectively, while 50.3% 60.0% improvement achieved method. These indicate approach significantly enhances efficiency has great potential address varying problems.