Modeling, Simulation and Control of Very Flexible Unmanned Aerial Vehicle

Abstract

This dissertation presents research on modeling, simulation and control of very flexible aircraft. This work includes theoretical and numerical developments, as well as experimental validations.

On the theoretical front, new kinematic equations for modeling sensors are derived. This formulation uses geometrically nonlinear strain-based finite elements developed as part of University of Michigan Nonlinear Aeroelastic Simulation Toolbox (UM/NAST). Numerical linearizations of both the flexible vehicle and the sensor measurements are developed, allowing a linear time invariant model to be extracted for control analysis and design. Two different algorithms to perform sensor fusion from different sensor sources to extract elastic deformation are investigated. Nonlinear least square method uses geometry and nonlinear beam strain-displacement kinematics to reconstruct the wing shape. Detailed information such as material properties or loading conditions are not required. The second method is the Kalman filter, implemented in a multi-rate form. This method requires a dynamical system representation to be available. However, it is more robust to noise corruption in sensor measurements.

In order to control maneuver loads, Model Predictive Control is applied to maneuver load alleviation of a representative very flexible aircraft (X-HALE). Numerical studies are performed in UM/NAST for pitch up and roll maneuvers. Both control and state constraints are successfully enforced, while reference commands are still being tracked. MPC execution is also timed and current implementation is capable of almost real-time operation.

On the experimental front, two aeroelastic testbed vehicles (ATV-6B and RRV-6B) are instrumented with sensors. On ATV-6B, an extensive set of sensors measuring structural, flight dynamic, and aerodynamic information are integrated on-board. A novel stereo-vision measurement system mounted on the body center looking towards the wing tip measures wing deformation. High brightness LEDs are used as target markers for easy detection and to allow each view to be captured with fast camera shutter speed. Experimental benchmarks are conducted to verify the accuracy of this methodology.

RRV-6B flight test results are presented. System identification is applied to the experimental data to generate a SISO description of the flexible aircraft. System identification results indicate that the UM/NAST X-HALE model requires some tuning to match observed dynamics. However, the general trends predicted by the numerical model are in agreement with flight test results.

Finally, using this identified plant, a stability augmentation autopilot is designed and flight tested. This augmentation autopilot utilizes a cascaded two-loop proportional integral control design, with the inner loop regulating angular rates and outer loop regulating attitude. Each of the three axes is assumed to be decoupled and designed using SISO methodology. This stabilization system demonstrates significant improvements in the RRV-6B handling qualities.

This dissertation ends with a summary of the results and conclusions, and its main contribution to the field. Suggestions for future work are also presented.