A collocation scheme for deep uncertainty treatment

Abstract

Considering an uncertain correlation length of the input random fields described by a Karhunen-Loève expansion leads to a probability-box approach for the stochastic finite element computation. But, these computations are highly costly. Then, a stochastic collocation method using sparse grids within a Smolyak algorithm is proposed to reduce the computational cost, particularly in the context of non-linear computations. The interest and the development of the Smolyak algorithm for stochastic model with non-linear finite element methods regarding mixed, aleatory and epistemic, uncertain inputs are here introduced. The limitations of Smolyak algorithm are critically discussed and suggestions for improvement are made.