Geometric modification for the enhancement of an airfoil performance using deep CNN

In the field of the flow control, the geometric modification is one of the flow control methods to improve the fluid dynamic performance of the structures. Thus, the present study utilizes a deep convolutional neural network (CNN) to modify various airfoil shapes to improve a lift to drag ratio (L/D). Because the surface pressure is predominant to L/D of bluff bodies, the surface pressure coefficient (Cp) distributions are considered as input to the CNN which will predict the corresponding airfoil shape. Hence, the present study numerically simulates viscous and steady 2D flow over the airfoil to obtain the accurate surface distributions of Cp as the dataset. Three approaches for the modification of the baseline Cp distributions are performed to achieve the favorable Cp distributions which are expected to giving higher L/D than the baseline ones. All flows around the predicted airfoils by the established CNN model are numerically simulated to achieve L/D and pressure fields. Finally, all predicted shapes give higher L/D and more preferable pressure fields than the corresponding baseline airfoils. Therefore, the present approach based on the deep CNN promises to the development of the geometric distur-bances as one of the flow control methods.

Automated summary

The study utilizes a deep CNN to modify airfoil shapes for improving L/D ratio, based on surface pressure distributions, and successfully predicts shapes with higher L/D and preferable pressure fields through numerical simulations and Cp distribution modifications.