Nowadays, artificial intelligence concepts are gaining popularity in many areas of engineering including fluid mechanics and computational fluid dynamics to minimize the computation cost. In the proposed study, artificial neural networks as an artificial intelligence approach are comprehensively investigated to predict aerodynamic coefficients of the RAE2822 airfoil, such as lift and drag coefficients. Database used in this study by the artificial neural networks is made by using SU2 computational fluid dynamics solver simulations under various freestream velocities and angle of attacks. This database is prepared to be able to train and test multi-layer perceptrons, which are capable of predicting intended values, lift and drag coefficients. During these processes, 85% of data are used to train multi-layer perceptron algorithms, whereas 15% of data are used for the test and the validation purposes. Hyperparameters used in this study are 30000 epochs, 100 batch size and 0.001 learning rate. As a result of the study, lift and drag coefficients, which are aerodynamic coefficients, are predicted with very high accuracy when velocity and angle of attack attributes are inserted into the multi-layer perceptron algorithm as inputs. By using multi-layer perceptron approach, correlation coefficients of lift and drag coefficients are found as 0.99426 and 0.97357, respectively. This is achieved as the best result within 12 different multi-layer perceptron architecture designs.
Anahtar Kelimeler: Multi-layer perceptron, Artificial Neural Network, Artificial Intelligence, SU2, Airfoil, Computational Fluid Dynamics
|
