Grid enabled optimisation using evolutionary algorithms

Optimisation and decision support tools are vital in all areas of engineering. Many engineering design problems, from the design of a controller for aircraft stability to the development of automobile chassis, can be effectively addressed by using evolutionary algorithms to optimise computational models of the systems under consideration. Unfortunately, for non-trivial problems, capturing the dynamics of a system with high fidelity often results in a model that is very computationally expensive. However, this level of fidelity is needed for an engineer to have confidence in the final solutions produced by an optimiser. Evolutionary algorithms aggravate this problem by often requiring many thousands of candidate solutions to beevaluated, since they search a population of points, before finding a satisfactory final solution. However, evolutionary algorithms do exhibit a large degree of parallelism, making them well suited for exploiting the emerging paradigm of Grid computing in which engineers and scientists have transparent access to large amounts of compute resources 'on demand'. In this work, strategies to accelerate the process of optimisation in engineering design are investigated. One promising approach examined in this thesis is the use of computational Grids in engineering design optimisation to reduce the time needed to obtain useful results. This allows the optimisation of much higher fidelity models than was previously possible. Another promising strategy is in the computational steering of multi-objective evolutionary search methods, where decision making is closely integrated into the search process resulting in a decision maker having finer control over the algorithm than was possible before.