Explainable nonlinear state-space modelling: analysing the GVT of a battery-operated aircraft

ifasd2024 Tracking Number 197

Traditionally, ground vibration tests (GVTs) of aircraft are processed using modal analysis algorithms. Most algorithms that are in commercial use today are grounded in the domain of linear system identification. These tools have proven their worth and are still the state of the art in commercial aviation, even though advances have been made in the research community. One of the more promising advances is the ability now to develop fully nonlinear models from experimental data. This capability could drastically improve the amount of information that can be extracted from a GVT. So far, non-linear effects, which are common in structural vibration of aircraft (freeplay of control surfaces, stick-and-slip behaviour at hinges, large deformations of slender wings, nonlinear friction,…), were considered a distortion of the underlying linear aircraft model that was sought. However, instead of trying to reduce the impact of the nonlinear distortion, we show the benefit of including nonlinearities in the model structure. In this work, we utilise recently developed concepts (a nonlinear state selection method, nonlinear function decoupling, a single-branch neural network) embedded in a nonlinear state space modelling framework, to build a nonlinear, explainable, data-driven model. We illustrate the methodology first on an analytical example containing multiple linear modes and a nonlinear distortion, and compare the performance of classical linear techniques to our nonlinear modelling framework. Then, we demonstrate this framework on a real-life multiple-input-multiple-output (MIMO) ground vibration test of the Magnus eFusion light sports aircraft.