Genetic Algorithms have been used in several engineering problems with clear advantages over other more traditional algorithms. The major advantages of these techniques are mainly related to robustness of the procedure. In simple GA applications however this major advantage is paid with computational cost: a large number of fitness evaluation is needed to reach a satisfactory solution.

If the fitness function is computed by means of complex simulation codes the total cost of the approach may make the problem impossible to be faced.

With the continuing growth of computing resources available the attention of engineers has modified the role of complex simulation that is more and more used directly in the design process. This aspect has also underlined the substantial weakness of traditional optimization approaches that can usually produce only single-objective optimized solution and only if the objective function satisfies continuity and often derivability conditions. This fact together with the need of multidisciplinary approach to design caused a growing interest into the use of Genetic Algorithms as general purpose optimizers. A large number of examples of engineering application can in fact be found in the literature [1].

Most real-life design procedures are complex tasks that have to deal with multidisciplinary environments, not always clearly defined targets, constraints to be satisfied. In this sense even though the target of the optimization could be expressed with a single expression like: "do the best possible design", the optimization process must consider several different usually conflicting objectives and the compromise obtainable might not be a-priori known. The possibility of looking not only for a single good solution but for a set of solutions (the "Pareto Set") [11] that satisfy different levels of compromise might be of great help to the decision maker that must select the most suitable one.

Three main issues makes however GA more than attractive and maybe unique among the aerodynamic design optimization methods: GA are usually much more robust than gradient based algorithm and can tolerate even approximate or noisy design objectives evaluation, GA can be efficiently parallelized and can therefore take full advantages of the massively parallel computer architecture, GA can directly approach a multi-objective optimization problem [8], [9], [10].

It must be noted however that the main concern related to the use of Genetic Algorithm for engineering problems involving the use of complex simulation codes is the computational effort needed for the accurate evaluation of a design configuration that, in the case of a crude application of the technique, might leads to unacceptable computer time if compared with other more classical algorithms [7]. With the help of parallel supercomputers [5], [6] and considering the fact that the computational performances of available machines is continuously growing, this problem at first glance might seem to be solvable by the computer technology development. However it is also known that the most powerful today available computer is still far from having sufficient performance even for single "multiphysics" simulation and therefore any effort in the direction of computational cost reduction of the optimization process should be seen at least as an opportunity to face more challenging design problems.

Two major paths exist in the reduction of computational effort:

The first can be faced by suitable operators the latter by the adoption of approximation techniques.