Parameter Estimation from Near Stall Flight Data using Conventional and Neural-based Methods

Abstract

The current research paper is an endeavour to estimate the parameters from near stall flight data of manned and unmanned research flight vehicles using conventional and neural based methods. For an aircraft undergoing stall, the aerodynamic model at these high angles of attack becomes non linear due to the influence of unsteady, transient and flow separation phenomena. In order to address these issues the Kirchhoff's flow separation theory was used to incorporate the nonlinearity in the aerodynamic model in terms of flow separation point and stall characteristic parameters. The classical Maximum Likelihood (MLE) method and Neural Gauss-Newton (NGN) method have been employed to estimate the nonlinear parameters of two manned and one unmanned research aircrafts. The estimated static stall parameter and the break point, for the flight vehicles under consideration, were observed to be consistent from both the methods. Moreover the efficacy of the methods is also evident from the consistent estimates of post stall hysteresis time constant. It can also be inferred that the considered quasi steady model is able to adequately capture the drag and pitching moment coefficients in the post stall regime. The confidence in these estimates have been significantly enhanced with the observed lower values of Cramer-Rao bounds. Further the estimated nonlinear parameters were validated by performing a proof of match exercise for the considered flight vehicles. Interestingly the NGN method, which doesn't involve solving equations of motion, was able to perform on a par with the MLE method.

Lateral directional parameter estimation of a miniature unmanned aerial vehicle using maximum likelihood and Neural Gauss Newton methods

“…A dedicated Graphical User Interface (GUI) has been developed using Lab-View to perform data logging as well as online display at the ground station. The flight test model is equipped with the above-mentioned data acquisition system consisting of a 9 degree of freedom (DOF) inertial measurement unit (IMU) capable of sensing linear accelerations (a x , a y , a z ), angular rates (p, q, r) and spatial orientation (φ, θ, ϕ) of the flight vehicle, absolute and differential pressure sensors, global positioning system (GPS) sensor, vane-type flow-angle sensors (α, β) and potentiometers to measure control surface deflections (δ e , δ a , δ r ) (29)(30)(31) . The flight velocity was obtained with the help of a differential pressure sensor attached to mini Pitot and static tubes which were fabricated in-house.…”

Determination of Parameters during Quasi-Steady Stall Maneuver Using Genetic Algorithm

“…The quasi-steady stall is convenient to perform but flight data gathered would enable to estimate only steady state stall characteristics,i.e., hysteresis time constant. Kumar and Ghosh [35][36] , Kumar and Ghosh et al[37] ,Sadrela and Dhyalan [38] employed the Kirchhoff's model of quasi-steady stall on flight test data gathered during Quasi-Steady Stall maneuver towards the estimation of steady-state stall characteristics and longitudinal aerodynamic parameters of HANSA 3 aircraft.…”

Parameter Estimation of Unstable Aircraft using Extreme Learning Machine

“…Except the earlier methods of estimating stability and control derivatives using ANN, the two methods namely, delta and zero have been reported in the literature using the concept of numerical finite difference approach [12][13][14] . An extension of these methods, the modified delta and Neural Gauss Newton (NGN) method have also been reported which yield the estimates with lesser standard deviations [15][16][17][18] . A radial basis function neural network can also be used for estimation of parameters as discussed in earlier methods 19 .…”