Prediction of lift forces of a non-slender delta wing under ground effect using Artificial Neural Network

Tümse S. , Bilgili M. , Şahin B.


  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Şanlıurfa
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.65


In this study, artificial neural network (ANN) method was applied to predict lift force of a non-slender delta wing under ground effect. The delta wing has a sweep angle of Ʌ=40° and chord length, c=14 cm. The variation of lift forces under the effect of the ground was revealed at angles of attack within the range of 2°≤α≤35° with 3° interval. In this investigation the variation of distance between trailing edge of the delta wing, h normalized with the chord length, c of the wing. The h/c which is the one of the most important parameters of the study was selected as h/c= 0.7, 0.55, 0.4, 0.25 and 0.1. The aim of the investigation is to predict lift force generated by the delta wing in case of h/c=0.4 by using actual lift forces in cases of h/c=0.7, 0.55, 0.25 and 0.1. Resilient propagation (RP) learning algorithm was carried out in the current simulation. The hidden layers and output layer of the network includes logistic sigmoid transfer function (logsig) and linear transfer function (purelin) as an activation function. At the end, lift force values acquired by artificial neural network (ANN) were compared with the actual values of lift forces. The mean absolute percentage error (MAPE) which is a measure of closeness of estimated value to the actual value was found as 2.1%. On the other hand, the two important parameters to assess the accuracy of estimated value, mean absolute deviation (MAD) which states the average of absolute error between predicted and real values and the root mean square error (RMSE) was calculated as 2.8% and 3.1%, respectively. It can be said that MAPE, MAD and RMSE are within the acceptable ranges. Thus, this study demonstrates that lift force of the delta wing under ground effect can be accurately predicted by using artificial neural network (ANN).