Robust shape optimization in aeronautics

Optimization is becoming an important field of research. The availability of more powerful computational resources, the ever-seek better performance and the company needs regarding time-to-market reduction, leaded to the necessity of better designs in less time. Aeronautics field has not been an exemption; shape optimization is a largely studied problem. It can be applied to many disciplines in this field, always intended to improve existing performances, and reduce costs and annoying characteristics like noise or consumption. Traditionally, optimization procedures were based on deterministic methodologies [6] so, it means that optimization leads to an optimum value which is an optimal point. Engineers have realized that this optimal point is no longer their aim. They need to ensure the optimal behaviour in the whole operating range; if they do not consider what happens in the vicinity of this point, a problem can arise. That is, in many cases, if the working point differs from the original, even a little distance, efficiency is reduced considerably [9]. Non deterministic methodologies have been applied to many fields [7][8], but the time consuming calculations required on CFD avoid to be applied to shape optimization. The study of the variability of the result against variability of the input parameters is a better representation of the real world; using these kinds of methodologies we can simulate from manufacturing tolerances to unknown measurement errors. The result is no longer a point, but a range or a set of values which defines the area where, in average, optimal output values are obtained. The optimal value could be worse than other optima, but considering its vicinity, it is clearly the best. This communication presents the development of a non-deterministic strategy coupled with optimization techniques. The main aim is the definition of an optimization procedure that ensure greater robustness of the solution than usual procedures. In order to reach this aim, probabilistic data will be used to define input parameters, and applied to each one of the members defined by the optimizer. In our case, we will use a genetic algorithm as optimizer