Low-Cost Wind Sensing for Dynamic Soaring UAVs.
Open Access
- Author:
- Quindlen, John Francis
- Graduate Program:
- Aerospace Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- July 27, 2012
- Committee Members:
- Jack Langelaan, Thesis Advisor/Co-Advisor
- Keywords:
- flush air data sensing
dynamic soaring
unmanned aerial vehicles - Abstract:
- Small UAVs have tremendous potential in military and civil applications, but are limited by their size in payload capacity and flight endurance. Extra batteries or fuel can be added to increase endurance, but at the expense of mission payload. Autonomous soaring techniques directly inspired by birds extract energy from the environment and offer the ability to increase UAV endurance without sacrificing payload. Dynamic soaring exploits naturally-occurring spatial wind gradients to gain energy and has been used by albatrosses to routinely fly hundreds of miles out at sea. This technique requires precise knowledge of the wind vector and albatrosses are the only birds capable of it due to specialized sensory organs in their beak nostrils. A flush air data sensing (FADS) system located on the nose of a soaring-capable UAV operates similar to an albatross beak and measures the aircraft's orientation with respect to the wind. The approach aims to create a small, versatile, low-cost sensing system in comparison to the large, expensive, commercial-grade wind sensors currently available. The design of the system was developed from potential flow and panel method models. Flush pressure ports drilled into the surface of a detachable nosecone feed airspeed, angle of attack, and angle of sideslip measurements to an Arduino microcontroller acting as a programmable air data computer. The entire FADS system cost less than $130 to construct and instrument, but still measures the complete wind vector. Neural networks are implemented to compute the wind vector. Wind tunnel tests collected training and validation data for airspeeds from 9 m/s to 27 m/s and aerodynamic angles from -15 to +15 degrees. Five neural networks of 3, 9, 15, 30 and 45 neurons were created for each wind parameter from the training data. Validation results confirm the more complex networks outperform the simpler 3, 9, and 15 neuron models, although not by considerably more. Ultimately, a 30 neuron network model is chosen as it performs just as well as the 45 neuron model, but with a simpler structure. Numerous flight tests conducted with a FADS-equipped sailplane demonstrate the capability of the system in real world flight regimes. Pilot-induced longitudinal and lateral-directional maneuvers verified the system's accuracy in dynamic maneuvers and steady glides. The preliminary flight test results prove the neural network-based FADS system's ability to accurately compute the wind vector for use in dynamic soaring research.