Year: 2026
Type: preprint
Abstract: This paper presents and validates a perturbation-based method in conjunction with feedforward neural network for estimation of aircraft stability and control derivatives. The method, named Double Delta Method, draws its inspiration from Taylor series expansion. Two sets of simulated data were generated using two different sets of flight control inputs of ATTAS aircraft by solving the lateral-directional equations numerically using Runge-Kutta 4th order method in MATLAB ® R2020a. First dataset, containing five motion/control variables at input nodes and one of the coefficients at output node at a time, is considered to train neural model. Thereafter, modified input files were created by adding small perturbation values of 0.001 and 0.002 in the motion variables and supplied to trained neural model for predicting the values of corresponding coefficient. For parameter estimation, the method uses modified input files and corresponding predicted values of corresponding coefficient. The robustness of the neural model and the method in estimating the parameters are also evaluated by adding the Gaussian noise of 10% in the simulated data. The estimates obtained from proposed method are also compared with the estimates obtained from existing Delta Method and Maximum Likelihood Method. For 10% noisy data, the CY derivatives obtained using Double Delta Method showed an improvement ranging from 4.5% to 41% compared with Delta Method, whereas the improvement compared with Maximum Likelihood Method ranged from 7% to 44%. A significant improvement, ranging from 0.4% to 25%, is also observed in the values of Cl and CN derivatives obtained from Double Delta Method. The second set of data is used for the validation of all three methods by comparing graphically the simulated and estimated motion variables and coefficients. The results demonstrate significant improvement over those obtained using the Delta method and Maximum Likelihood Method.
Source: Research Square
Institution: Amity University
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
Funder: Amity University
OpenAlex ID: W7226019469
