iFlyNet: Inferring UAV Flight from Wing Behavior

Abstract

We introduce iFlyNet, a novel flight state awareness system that is robust, lowprofile, and more reliable in harsh conditions than conventional methods. The system interprets the global flight state of a UAV by monitoring multimodal structural characteristics of its wings, where aerodynamics are predominant. A microfabricated transducer array consisting of temperature compensated strain and piezoelectric sensors captures the wing’s static and dynamic stress profiles at multiple spatial locations. This data informs our deep one-dimensional convolutional neural network-based sensor fusion algorithm designed to run on flight vehicle computer at near real-time speeds. Wind tunnel experiments with the sensors installed on a single wing demonstrate that iFlyNet can accurately predict the flight state variables of airspeed and angle of attack across the flight envelope of the UAV. In addition, we also show the benefits of our system compared to status-quo techniques through accurate predictions of lift and drag forces and high-precision tracking of stall occurrence at highly variable environmental conditions. By nearly matching the readings of traditional state measurement devices at a fraction of weight as well as providing accurate predictions of flight performance and safety-critical metrics, our system offers a unique paradigm for aircraft state identification.