Panday, Aarti2024-01-232024-01-232024https://hdl.handle.net/10539/37384A thesis submitted in partial fulfilment of the requirements for the degree Doctor of Philosophy to the Faculty of Engineering and the Built Environment, School of Mechanical, Industrial & Aeronautical Engineering, University of the Witwatersrand, Johannesburg, 2022Unmanned aerial vehicles (UAVs) have evolved in design, complexity and have seen increasingly diverse roles within the past five decades of its adoption as an essential aviation tool. The lack of an in-flight-refuelling capability is one of the major practical limitations of UAVs. In a successful automated aerial refuelling (AAR) manoeuvre, the refuelling receiver aircraft must maintain an accurate position relative to the tanker aircraft. Without the presence of a skilled pilot in the cockpit, control systems must be developed to meet the stringent performance and safety requirements of the AAR manoeuvre, while requiring stability and performance robustness. The aim of this study was to address the development of stable control laws by synthesising techniques in nonlinear and intelligent control. The objectives included system model development, optimisation of controller designs using evolutionary algorithms coupled with Lyapunov stability theory, and system simulation both with and without disturbances. The research objectives were addressed in three phases using numerical simulation in the MATLAB environment: establishing a baseline case in phase 1 where proportional-integral-derivative (PID) control was used, followed by the development of a Lyapunov-based feedback linearisation controller in phase 2 and lastly, the development of a dynamic neural network-based control scheme in phase 3. An aerial refuelling system model was developed, inclusive of the tanker-receiver engagement, where the receiver aircraft dynamics was framed as a variable-mass system subject to a wind field as inherent to the process. Three controllers were designed for the system. The nonlinear system model was used directly in the development of a benchmark controller case using PID control for the aerial refuelling manoeuvre. A single set of controller gains, applicable throughout all phases of the refuelling manoeuvre, was as found through manual tuning, eliminating the need for gain scheduling. Lyapunov stability theory was extended to the aerial refuelling system, where mechanical energy of the variable-mass system was increased as fuel was added to the receiver aircraft. Stability was established for the system throughout refuelling, including the post-refuelling phase, where the receiver aircraft would remain on station and mechanical energy would be conserved. The stability theorem was incorporated into a cost function, whose minimisation using evolutionary algorithms, would lead to optimisation of controller gains that would ensure system iv stability. The manually-tuned controller gains for the benchmark PID controller were optimised using particle swarm optimisation and genetic algorithms (GAs). It was found through the minimisation routine, that there is a minimum energy threshold that must not be exceeded in the system for performance and stability to be maintained. The minimum energy was found to be related to the ratio of mass increase to total aircraft mass. Two disturbance cases were defined: decreased, asymmetric refuelling and turbulence, to test the robustness of the designed controller. The second controller designed was based on feedback linearisation (FBL). FBL was incorporated in the inner loop, while a pseudo-backstepping scheme in the command inversion loop converted the reference command signals into desired dynamics. PID tracking control was used to close the outer loop to give effect to controller tracking objectives. Controller gains were selected through manual tuning and were then optimised with the Lyapunov-based GA routine developed earlier. The third controller design was the synthesised nonlinear intelligent controller, which used dynamic neural networks (DNNs) in an indirect adaptive FBL control scheme in the inner loop, while the outer loop remained the same as the FBL controller without the DNN. System identification was performed in order to obtain the DNN models to estimate the plant dynamics. Controller gains were selected through manual tuning and were then optimised with the Lyapunov-based GA routine developed earlier. Dynamic simulation results showed that the system was sensitive in the lateral-directional states in the presence of disturbances. The benchmark PID control was the least robust to disturbances, but its performance was improved when the Lyapunov-based GA gains were applied. The manually-tuned FBL controller exhibited damped oscillatory behaviour in the lateral-directional states during refuelling, however, application of the Lyapunov-based GA gains led to these behaviours being eradicated from the system state-variable trajectories. The DNN-based FBL controller with Lyapunov-based GA-optimised gains performed the best from all designed controllers. Transient behaviour with maximum deviation of 0.004% from nominal in the longitudinal and vertical separation was observed in the DNN-based FBL controller prior to refuelling commencing, but this very quickly settled to nominal. Application of the Lyapunov-based optimisation gave improved controller performance compared to the manually-tuned control gains in all instances, and the system was able to return to its nominal or desired values when subjected to disturbances. For the AAR manoeuvre, the synthesis of nonlinear and intelligent control techniques was able to produce stable control which showed good robustness characteristics when subjected to disturbances.enUnmanned aerial vehicles (UAVs)Automated aerial refuelling (AAR)Thesis