Optimization involves representation of a problem in terms of a vector of variables and maximization or minimization of one or more objective function(s) subject to a set of constraints. While efficient algorithms to identify the optimal solution exist for a number of important classes of problems (linear programming, quadratic programming etc), there is a host of real world optimization problems where locating the optimal solution is not trivial.

Fortunately for such problems, it is sufficient to be able to locate a good solution. Heuristic methods are particularly attractive for such problems as they require no assumption about the nature of the function or its slope and at the same time, they are easy to implement.

There has been significant work over the years to develop numerous heuristic approaches (genetic algorithms, evolutionary algorithm, swarms, differential evolution etc). However, all such approaches can deal with problems with a fixed number of variables. Although, such approaches have resulted in excellent benefits for a number of real life problems, they fall flat for a class of applications where there is a need to explore across multiple concepts i.e. problems where individual solutions may be represented by different length variable vectors.

In this PhD project, we aim to explore solution representation using genetic programming where commonalities between designs can be exchanged and complex designs can evolve through growth and inheritance. Our intention is to represent designs using trees and exchange of components and features between designs based on statistical significance. Estimation of Distribution Algorithms (EDA) will be employed to identify inherent hidden s tructures to create potential designs and variants. The approach will provide a means to solve a wide class of optimization problems which include identification of number and capacity of warehouses for a distribution network, identification of an optimal design where the design space spans across multiple concepts, identification of number of layers and their thickness and constituents for a composite laminate design or even to solve a variety of industrial engineering problems where solutions may vary in dimensionality. Additional challenges will involve inheritance of useful components under a multiobjective scenario where nondominance needs to be considered. The methodology also needs to be computationally efficient for real life implementation.

Essential Skills: Good coding skills in C/C++, MATLAB Preferred Skills: Exposure to and use of CAD Tools such as Pro-E/CATIA/Solidworks would be useful.

Requirement: Must meet UNSW PhD Admission Requirements and should be able to join in Session 1, 2007.Contact:
Dr. Tapabrata Ray
School of Aerospace, Civil and Mechanical Engineering
University of New South Wales,
Australian Defence Force Academy Northcott Drive,
ACT 2600
Email:
Tel: 612-62688248

Supervisors: Dr. Tapabrata Ray (UNSW@ADFA), Dr. Warren Smith (UNSW@ADFA), Dr. Stuart Cannon (Defence Science and Technology Organization (DSTO, Australia)

Last Date: January 15, 2007. The search might be terminated early if an appropriate candidate is found earlier.