The proposed optimisation procedure to design complex mechanical systems takes several steps: Formulation stage The designer has to provide validated mathematical models (physical models) and find a suitable set of design variables and performance indexes which will be taken into account during the optimisation phase. Explorative stage The physical model (or group of models) is used in this stage to compute the relationships between design variables and performance indexes. This is done for a "Uniformly distributed sequence" of feasible design variables vectors. The analysis is performed by using the Spearman rank correlation coefficient. If a strong correlation between performance indexes is found, the number of performance indexes that have to be taken into account can be reduced. This phase is important because if the designer is not satisfied with the preliminary results, he can go back and reformulate the problem changing the design variables set and the performance indexes set. Search stage When the problem is relatively simple, the optimisation can be performed directly using the physical model. Approximation models (fitting the response data given by the physical model) should be used in optimisation when the computational effort for optimising the physical model is prohibitive. Several approximation methods can be considered (Artificial Neural Networks, Kriging, Polynomial Approximation). The Pareto-optimal set can be computed in a very short time by resorting to the Approximation model. Analysis and Synthesis A second correlation analysis is performed on the solutions belonging to the Pareto-optimal set. This can be done using again the Spearman rank correlation coefficient. The selection of one preferred solution among the Pareto-optimal ones constitutes the synthesis stage.

---