Aircraft virtual sensor design via a Time-dependent Functional Pooling NARX methodology

A Time-dependent Functional Pooling Nonlinear AutoRegressive with eXogenous excitation (TFP-NARX) methodology for aircraft virtual sensor design is introduced. The methodology is unique in establishing a single but global nonlinear dynamical model, referred to as TFP-NARX, from data obtained from several individual flights, each corresponding to different flight conditions and aircraft configurations. This model structure is demonstrated to be capable of accurately representing the aircraft nonlinear dynamics under various flight conditions and aircraft configurations in terms of aircraft weight and center of gravity. Based on this model structure, three individual designs are postulated: A “fixed” design pertaining to a single aircraft weight and center of gravity configuration, as well as two more involved designs, a “basic” and an “extended” one, that account for various aircraft configurations. The three designs are subsequently employed for the development of angle-of-attack virtual sensors for the main flight regimes (landing, take-off, clean flight) of a small commercial aircraft. The performance of the developed virtual sensors is examined via a number of validation flights within a nonlinear simulation environment and are shown to meet the requirements of low mean and peak error.