The present Doctoral thesis is about System identification of a flexible wing aircraft, taking into consideration the effect of structural flexibility over the aircraft rigid body responses. The flexibility imposes additional degrees of freedom at the system and its effect also generates additional aerodynamic coefficients that influences the aircraft movement at the 6 DoF of rigid body motion. There are methodologies that can physically describe this structural-aerodynamic coupling. However, they are very complex, and difficult to model and even more difficult to perform identification. Because of that, Data Driven identification is proposed at the present work, to identify a more accurate model for the system. Three different techniques of Data Driven identification are used here, all of them based on different types of Deep Neural Networks: The Feed Forward Neural Network (FFNN), the Long Short-Term Memory Neural Network (LSTM) and the Physics Informed Neural Network (PINN). The main contribution of the present work is the use of Physics Informed Neural Network, which uses Ordinary Differential Equations as information for the loss function to make the Neural Network to converge faster and to have a better accuracy in terms of results. Because PINN uses Ordinary Differential Equations, the physical ODEs that describes the aircraft equations of motion with flexibility embedded are used to inform the loss function of the aircraft, equations that were used at model-based identification of flexible aircraft. The aircraft used for the present work is an UAV called EOLO, which has long wingspan and high flexible wing, which has embedded Inertial Sensors, Air data sensors, accelerometers and strain gauges. Specific Flight Test maneuvers were chosen to be performed to excite aircraft rigid body responses. Most of the maneuvers were performed with automatic scripts programmed at the pilot controller, to have better results. Results shown that the FFNN network can obtain better accuracy than LSTM, because FFNN can fit better high dynamics/noise. However, the PINN can give a more physical meaning to the results, even if PINN did not prove to have better accuracy and/or to converge faster.