Efficient Decoupling of Multiphysics Systems for Uncertainty Propagation

Abstract

Uncertainty propagation through coupled multiphysics systems is often intractable due to computational expense. In this work, we present a novel methodology to enable uncertainty analysis of expensive coupled systems. The approach consists of offline discipline level analyses followed by an online synthesis that results in accurate approximations of full coupled system level uncertainty analyses. Coupling is handled by an efficient procedure for approximating the map from system inputs to fixed point sets that makes use of state of the art L1-minimization techniques and cut high dimensional model representations. The methodology is demonstrated on an analytic numerical example and a fire detection satellite system where it is shown to perform well as compared to brute force Monte Carlo simulation.