A response surface method based on support vector machines trained with an adaptive experimental design

Structural reliability is nowadays largely used to take into account uncertainties related to the input data of a structural model. When the structural response becomes complex, the mechanical model is designed within the framework of the finite element method and therefore, the computational time required by a coupling reliability/finite element analysis is driven by the number of performance function calls. This paper aims at proposing an original approach to approximate implicit limit state functions. It is based on the support vector machine used in regression trained with an adaptive experimental design. Several numerical examples proposed in the published literature are considered to assess the efficiency of the proposed method.