Data-driven stochastic identification for fly-by-feel aerospace structures: Critical assessment of non-parametric and parametric approaches

摘要

In this work, a comparison and assessment of data-driven non-parametric and parametric stochastic identification approaches is presented. The non-parametric Welch-based spectral estimation and parametric "global" time-series modeling approaches are presented and assessed via a series of wind-tunnel experiments under varying flight states. In this context, the term "global" refers to the identification of a model that is capable of representing the structure under any admissible flight state based on data recorded from a sample of these states. The global model identification framework is based on stochastic time-series models for representing the structural dynamics and aeroelastic response under multiple flight states, with each state characterized by several variables, such as the airspeed, angle of attack, altitude and temperature, forming a flight state vector. In addition, based on these identification approaches, two novel flight state identification approaches are postulated and experimentally validated, namely a non-parametric stall detection approach based on the statistical signal energy analysis and an inverse-type machine-learning method leveraging the use of stochastic global time-series models. The experimental evaluation and assessment is based on a prototype bio-inspired self-sensing composite wing that is subjected to a series of wind tunnel experiments under multiple flight states. Distributed piezoelectric sensors embedded in the composite layup provide the sensing capabilities. Experimental data collected are employed for the non-parametric and parametric identification approaches, with the latter being based on appropriate parameter estimation and model structure selection methods. The identified models are able to successfully represent the wing's aeroelastic response under the admissible flight states via a minimum number of estimated parameters, while the stall detection and machine-learning flight state identification approaches constitute a first step towards the next generation of "fly-by-feel" aerospace vehicles with state awareness capabilities.