Data-Driven State Awareness for Fly-by-Feel Aerial Vehicles via Adaptive Time Series and Gaussian Process Regression Models

This work presents the investigation and critical assessment, within the framework of Dynamic Data Driven Applications Systems (DDDAS), of two probabilistic state awareness approaches for fly-by-feel aerial vehicles based on (i) stochastic adaptive time-dependent time series models and (ii) Bayesian learning via homoscedastic and heteroscedastic Gaussian process regression models (GPRMs). Stochastic time-dependent autoregressive (TAR) time series models with adaptive parameters are estimated via a recursive maximum likelihood (RML) scheme and used to represent the dynamic response of a self-sensing composite wing under varying flight states. Bayesian learning based on homoscedastic and heteroscedastic versions of GPRM is assessed via the ability to represent the nonlinear mapping between the flight state and the vibration signal energy of the wing. The experimental assessment is based on a prototype self-sensing UAV wing that is subjected to a series of wind tunnel experiments under multiple flight states.