Soft computing in engineering design optimisation-optimiser behaviour evaluation through stochastic analysis

A systematic and computationally efficient method for the global shape optimisation of propellers is outlined. Soft Computing methodologies are used extensively throughout the framework to enable the solution of this complex multiple-objective and multi-disciplinary problem. These technologies include Simulated Annealing for design optimisation, Neural Networks for analytical system and experimental data representation, heuristics for the handling of objectives and constraints in order to guide the optimiser to a feasible solution, and adaptive function evaluation to decrease the computation time of the objective function. The focus of this paper is the interpretation of optimiser behaviour using stochastic analysis of resulting designs to show the effectiveness of applying these technologies to this problem. The framework is show to succeed in producing designs significantly improving on both their noise and performance requirements.