Development of Airfoils Based on Their Aerodynamic Characteristics Using Artificial Neural Networks

One of the main concerns in the development of aerodynamic airfoils is efficiency. This variable, which corresponds to the ratio between aerodynamic force and drag (force contrary to the aircraft movement), is highly important to the aircraft performance. The higher is the efficiency value, the lower is the aircraft fuel consumption, although an increase in resistance commonly follows the increase in lift, nevertheless, it is still intended to project new airfoils to match these features. Following this line of thought, there are many works in literature (Eppler, 1974; Truckenbrodt, 1951; Wortmann, 1961) applying techniques for airfoil design, and also among these techniques, artificial neural networks are used (Rai and Madavan, 2000). Ross et al. (1997) used artificial neural networks to minimize the amount of necessary data to completely define the aerodynamic performance of a model examined in a wind tunnel. In this case, the author experimentally used only 50% of the data, and achieved results with the trained neural network which were very close to the ones from the performed measurements. Wallach et al. (2006), Rajkumar and Bardina (2002, 2003) and Soltani et al. (2007) used other applications of the ANN in airfoils, in the prediction of aerodynamic coefficients of airfoils and aircrafts using artificial neural networks. Norgaard et al. (1997) used ANN for more efficient design evaluations and in order to find ideal configurations for flaps. The author also used four networks to predict the CL, CD, CM and L/D coefficients and one network to analyze the flap configurations. On the other hand, Maezabadi et al. (2008) analyzed the air draining in a section of a blade of a wind turbine, using the network to predict the behavior of the flow in the airfoil for the desired conditions. Nowadays, there is a lot of software able to display the aerodynamic behavior of an airfoil from its geometry. Such methodology is opposed to the proposition of this work, which is to develop the airfoil (geometry) from some of its important aerodynamic attributes. This work aims to develop airfoils from the desired characteristics (lift, drag and maximum efficiency rate), using an artificial neural network. Thereunto, an algorithm is used based on artificial neural networks with several different types of architecture. As specific goals, it stand out the evaluation of neural architectures presenting the lowest margin of error in relation to the airfoils used, and the development of new architectures based on committee machines (modular networks) aiming to improve results and become appropriate for the evaluated airfoils. It is worth mentioning that the airfoils are not used exclusively for application in aircrafts, but may also present other uses, such as wind power generator blades, pumps, fans, etc. ISSN 2176-5480