A fault Hiding Approach for Fault Tolerant Control of a Class of VTOL Vehicles

Abstract

The octorotor is an unmanned VTOL capable vehicle with eight motors with xed pitch rotors. It is controlled by varying the speeds of its eight motors which are placed around the vehicle. There is no need for a complex swashplate system, making the vehicle low cost and dynamically simple. The increase in the number of e ectors over the quadrotor allows for inbuilt hardware redundancy. It is this redundancy which is of particular interest as the capabilities and applications of VTOL capable UAVs increases and the payloads become more expensive and sensitive. It would be unacceptable for a hardware failure to result in the loss of the vehicle and payload, especially if operating in close proximity to people. An operational requirement is that the operator must be able to control the vehicle's position and yaw angle. Position reference commands are generated in an inertial frame and these must be related to the vehicle- xed frame through a rotation matrix. The downfall of this method is that trigonometric singularities exist for large body angles where gimbal lock can occur. For this reason the unit quaternion attitude representation is used. The octorotor is not open-loop stable so a PD controller is used to provide for singularity-free, almost global asymptotic stability which is capable of following ightpaths as well as recovering from an initial inverted attitude. The output of the controller is called the virtual control since this demand is passed to the control allocation subsystem where the overall forces and moments are generated. A suitable control allocation method is needed since there are more e ectors than actuated degrees of freedom. The e ectors are assumed to be linear and various methods are used to provide constrained control allocation. If the virtual control is constrained then the allocation problem is always the unconstrained allocation problem and is guaranteed to be successful. By applying the constraints directly to the e ectors it is not necessary to use complex face searching algorithms to calculate the constrained virtual control. An objective of this thesis is to ensure that e ector failures do not a ect the vehicle's

ight performance. This is integral to the aim of demonstrating that the hardware redundancy is su cient to allow ights over populated areas. E ector failures are modelled as an instantaneous loss of thrust from an e ector. This causes an adverse roll, pitch, and yaw disturbance as well as a drop in altitude. The recovery is based on the fault hiding method where the virtual control remains invariant from the nominal case and the fault is hidden in the plant. If none of the remaining e ectors are saturated then the failure-free performance is maintained and the operator should not notice any change to the vehicle handling. Kalman controllability analysis is done to determine the combinations of e ector failures which result in a controllable vehicle. Flight testing has demonstrated the suitability of the controller to the task of stabilising the vehicle. The failure scenarios are initialised before the ight and the performance is invariant to the health of the e ectors. The reasons for di erences between the simulation data and ight data are explained. Future work will implement an online fault detection scheme.