Evolutionary Design of Robust Flight Control for a Hypersonic Aircraft

An evolutionary design approach is used to construct an autopilot for a hypersonic airbreathing aircraft. Flight control for this class of vehicle is an extremely challenging problem due to the combination of nonlinear dynamics, parametric uncertainty and complex constraints. Consequently, simultaneous control over the flight path, aerodynamic attitude and propulsion is required. This thesis develops and applies a design procedure which can explicitly address the challenges of hypersonic flight control. The principal computational results of this thesis focus on the capability of an evolutionary based optimizer to design, without a priori knowledge, a robust fuzzy control law for a hypersonic vehicle concept. This work is not meant as an expression of the superiority of a particular control approach or an optimization procedure. Rather, it experiments with the potential of fuzzy control to represent a complex, nonlinear, and robust control function, the incorporation of robustness features in the control performance measure, and the capability of the genetic algorithm as a search procedure. The structure of the fuzzy rule base defines the mapping procedure and the design procedure learns the output profile through a numerical optimization procedure. The evolution of the controller design requires the definition of a scalar objective function which assesses the merit of the particular control solution being tested. For this work the design objective is extracted from a collection of simulated flight responses. Such an approach is computationally demanding, but the benefits are that fewer simplifying assumptions are required in the flight dynamics and aero-propulsive models. There is also the capacity to represent features in the objective function which encourage the development of a robust control law. These include the evaluation of the flight response at many points along the trajectory, the full range of expected attitude and control states, and the inclusion of realistic variations in engine operation, vehicle aerodynamics, and physical properties. Essentially, the controller can be configured based on the best available and most practical model of the system. Stability and performance robustness are therefore a natural derivative of the design exposure to the varied performance of the system. A conventional autopilot structure has been used for the longitudinal motion study. An outer guidance loop provides vehicle attitude commands for trajectory maintenance, while an inner-loop attitude controller tracks the commanded attitude and provides stability augmentation. The control design focus is on the specification of the control function for v the inner-loop. Aside from the evolutionary based design, the second prominent feature of the control application is the parameterization of the attitude controller through a fuzzy rule base. A fuzzy controller has been used for its inherent robustness, and its simplicity in representing a nonlinear control function. The main performance benefit over a constant gain linear controller is derived from the capacity to locally manipulate the control surface of the fuzzy controller during the design. This allows rapid attitude response while still providing the appropriate control authority about the trimmed condition. In addition, the control surface can be configured to any nonlinearities which are a function of the control inputs. The development of the flight simulation models and the control design procedure are described in detail in the thesis. For the flight simulator, particular attention was paid to a realistic representation of the flight dynamics behaviour through an aero-propulsive simulation module and a dynamics formulation that used the full six degree-of-freedom equations of motion for flight about a spherical, rotating Earth. Successful application of the evolutionary control design procedure to the hypersonic vehicle is demonstrated through a series of design experiments. These cover some of the many variants available with both the specification of the control function and the application of the genetic algorithm. Within this scope, the benefits and potential pitfalls of the overall procedure are considered. Significantly, the genetic algorithm is able to capture the necessary control features for a design of large dimension, with relatively few function evaluations. To provide guarantees of performance and stability robustness, the fuzzy controller must be assessed against an extensive set of test conditions throughout the design process. As part of the numerical experiments it was found that to achieve good quality control designs, a modification to a well know non-uniform mutation operator was required. This minor enhancement to the genetic algorithm greatly improved the quality of the control solution. The search performance benefits have also been demonstrated on a collection of standard minimization test problems, as documented in Appendix A. List of Publications [1] Austin, K. J. and Jacobs, P. A., “A Newtonian solver for hypersonic flows,” Technical Report 5/96, Department of Mechanical Engineering, The University of Queensland, 1996. [2] Smith, A. L., Johnston, I.A., and Austin, K.J., “Comparison of numerical and experimental drag measurement in hypervelocity flow,” The Aeronautical Journal, November 1996: pp 473-480. [3] Austin, K. J. and Jacobs, P. A., “Can trained monkeys design flight controllers for hypersonic vehicles,” The 10th Biennial Computational Techniques and Applications Conference, Brisbane, Australia, 2001. [4] Austin, K. J. and Jacobs, P. A., “Application of genetic algorithms to hypersonic flight control,” Joint 9th IFSA World Congress and 20th NAFIPS International Conference, Vancouver, Canada, 2001. [5] Stewart, B., Morgan, R. G., Jacobs, P. A., Austin, K. J., and Jenkins, D. M., Establishment of test conditions in the RHYFL-X facility, 37th AIAA/ASME/SAE/ASEE Joint Propulsion Conference and Exhibition, AIAA Paper 2001-3678, 2001. [6] Austin, K. J. and Jacobs, P. A., “Numerical flight Simulation of a hypersonic airbreathing vehicle,” Technical Report 2002/3, Department of Mechanical Engineering, The University of Queensland, 2002. [7] Austin, K. J. and Jacobs, P. A., “A real-coded genetic algorithm for function optimization,” Technical Report 2002/4, Department of Mechanical Engineering, The University of Queensland, 2002. [8] Austin, K. J. and Jacobs, P. A., “Benchmark Control Problem Evolutionary Design of a Robust Controller,” Technical Report 2002/5, Department of Mechanical Engineering, The University of Queensland, 2002.