Systematic optimal design of controlled systems using evolutionary computation

This project will consider the use of evolutionary algorithms in the computer aided design of controlled systems. In the optimal design of systems, computer simulations are often used as a “black box” that returns a performance estimate for a given set of design parameters that can include structural variables and controller gains. The more reliable the estimate (i.e. the lower the uncertainty in the performance measure for a given set) the longer the computation typically takes, leading to a tradeoff in how ‘noisy’ performance measures can be used for both speed and reliability of the decision space. In this project, methods of evolutionary design will be developed that rigorously address this tradeoff, ultimately leading to results that will provide faster convergence of the design to a specified accuracy of the local or global optimum solution. The methods will be applied to one current industry application involved with the co-design of model based controllers and structures.

The successful applicant will spend a period of their candidature at the University of Birmingham developing methods of analyzing the performance of evolutionary algorithms, and a period at the University of Melbourne establishing theoretical results for model based controller optimization.