Application of Evolutionary Strategies to Optimal Design of Multibody Systems

This paper is concerned with the application of evolutionary strategiesto the optimization of the kinematic or dynamic behaviour of mechanicalsystems. Although slower than classical deterministic, gradient-basedmethods, they represent an interesting alternative: they are global,they should be more robust as they do not rely on continuity andderivability conditions and they can use the simulation software as is.They are inspired from natural evolution: the design variablescorresponding to the genes of an individual mutate from generation togeneration and the ones who survive are those that are the best fittedto their environment, that's to say with the best objective function.The implementation of evolutionary strategies is presented as well assome guidelines to choose the most important parameters. Two examplesare developed for the sake of illustration: the kinematic optimizationof a suspension and the dynamic optimization of the comfort of a railwayvehicle. Finally, the performances are compared with respect todeterministic methods and genetic algorithms.