Abstract
For many optimum design problems, the cost function is the result of a complex numerical code, and the desired optimum is global. The global optimization code requires many evaluations of the cost function, which implies that the optimum cannot be obtained in a reasonable computation time. The first aim of this paper is to propose, on an example problem, a way of solving such complexity. The problem considered here concerns an aerospace vehicle which is intended to fulfil a given task. The control of the vehicle for achieving the task is obtained as the solution of an optimal time control problem. Our aim is to design the architecture of the vehicle so that it can complete the task in minimum time. For a given architecture the minimum time and the control are calculated using a very fast semi-analytical method, obtained by asymptotic matching. This minimum time is taken as the cost function for a given architecture design. The best architecture is then computed by a stochastic genetic-like global optimization algorithm. A presentation of this genetic algorithm is the second aim of the paper.