%0 Journal Article %T A Gray-Box Non-Parametric Aircraft System Identification Method Using the ANFIS Network for Prediction of High Angle of Attack Maneuvers %J AUT Journal of Modeling and Simulation %I Amirkabir University of Technology %Z 2588-2953 %A Bagherzadeh, Seyed Amin %A Mohammadkarimi, Hamed %D 2021 %\ 12/01/2021 %V 53 %N 2 %P 255-274 %! A Gray-Box Non-Parametric Aircraft System Identification Method Using the ANFIS Network for Prediction of High Angle of Attack Maneuvers %K gray-box model %K non-parametric model %K aircraft system identification %K ANFIS %K high angle of attack %R 10.22060/miscj.2022.20612.5259 %X This paper aims to identify a gray-box non-parametric model for the airplane nonlinear aerodynamics throughout high angle of attack maneuvers using the Adaptive Neuro-Fuzzy Inference System (ANFIS). The gray-box modeling is employed in this paper in which the force and moments are predicted rather than the flight parameters. Flight test data of a large-scale unpowered model of a fighter airplane is modeled by ANFIS, and the results are compared with the traditional multi-layer feed forward Artificial Neural Network (ANN). The employed gray-box identification method considers both the nonlinearity and the longitudinal-lateral/directional coupling effects. The control commands and the flight conditions are the inputs to the system identification block while the force and moment coefficients are the targets. The optimal values for the ANFIS parameters are adjusted by the hybrid learning algorithm in order to minimize the Mean Squared Error (MSE) between the best estimated, target force, and moment coefficients while the ANN is trained by several learning algorithms. The precision of the model is checked during the training and test phases for a single flight condition. Afterwards, the generalization of the model is checked for flight conditions dissimilar from the training one. The results indicate that the ANN has moderate precision in the test phase while the ANFIS has excellent precision. Furthermore, based on the results, the ANN predictions cannot follow the flight data in flight conditions dissimilar from the training ones while the ANFIS seems quite robust in those conditions. %U https://miscj.aut.ac.ir/article_4660_c96e68d6f3b555da24c06dab267727ca.pdf