Modeling and System Identification of an Ornithopter Flight Dynamics Model

Abstract

Ornithopters are robotic flight vehicles that employ flapping wings to generate lift and thrust forces in a manner that mimics avian flyers. At the small scales and Reynolds numbers currently under investigation for miniature aircraft, where viscous effects deteriorate the performance of conventional aircraft, ornithopters achieve efficient flight by exploiting unsteady aerodynamic flow fields, making them well-suited for a variety of unmanned vehicle applications. Parsimonous dynamic models of these systems are requisite to augment stability and design autopilots for autonomous operation; however, flapping flight is fundamentally different than other means of engineered flight and requires a new standard model for describing the flight dynamics. This dissertation presents an investigation into the flight mechanics of an ornithopter and develops a dynamical model suitable for autopilot design for this class of system.

A 1.22 m wing span ornithopter test vehicle was used to experimentally investigate flapping wing flight. Flight data, recorded in trimmed straight and level mean flight using a custom avionics package, reported pitch rates and heave accelerations up to 5.62 rad/s and 46.1 m/s^2 in amplitude. Computer modeling of the vehicle geometry revealed a 0.03 m shift in the center of mass, up to a 53.6% change in the moments of inertia, and the generation of significant inertial forces. These findings justified a nonlinear multibody model of the vehicle dynamics, which was derived using the Boltzmann-Hamel equations. Models for the actuator dynamics, tail aerodynamics, and wing aerodynamics, difficult to obtain from first principles, were determined using system identification techniques with experimental data. A full nonlinear flight dynamics model was developed and coded in both MATLAB and FORTRAN programming languages.

An optimization technique is introduced to find trim solutions, which are defined as limit cycle oscillations in the state space. Numerical linearization about straight and level mean flight resulted in both a canonical time-invariant model and a time-periodic model. The time-invariant model exhibited an unstable spiral mode, stable roll mode, stable dutch roll mode, a stable short period mode, and an unstable short period mode. Floquet analysis on the identified time-periodic model resulted in an equivalent time-invariant model having an unstable second order and two stable first order modes, in both the longitudinal and lateral dynamics.