Neural Network Techniques to Solve a Model-Free Scheme for Flow Angle Estimation

In the area of synthetic and smart sensors for flow angle estimation for UAS applications, model-based, data-driven and model-free approaches represent the state-of-the-art techniques to estimate the angle-of-attack and angle-of-sideslip. Thanks to sensor fusion techniques, a synthetic sensor is able to provide estimation of flow angles without any dedicated physical sensors. A model-free approach, based on a set of nonlinear equations, demonstrates good performances when used with flight simulated data. As shown in this work, an iterative solver could not be adequate dealing with real flight data affected by common instrument uncertainties. In order to cope with real flight data, two deterministic solvers can be adopted that are based on neural techniques: pre-trained multilayer perceptron and generalised radial basis function neural networks. The neural networks considered in this work are trained with batch and sequential algorithms. All solvers are tested with noise-free and noisy signals simulating real flight instrument noise. The aim of the present work, in fact, is to provide a preliminary benchmark between the aforementioned solvers when used to solve the proposed nonlinear scheme for flow angle estimation dealing with instrument noise.