Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
499087 | Computer Methods in Applied Mechanics and Engineering | 2009 | 9 Pages |
We develop a stochastic functional representation that is adapted to problems involving various forms of epistemic uncertainties including modeling error and data paucity. The new representation builds on the polynomial Chaos decomposition and eventually yields a Karhunen–Loeve expansion with random multiplicative coefficients. In this expansion, one set of uncertainty is captured in the usual manner, as uncorrelated scalar random variables. Another component of the uncertainty, statistically independent from the first, is captured by constructing the, usually deterministic, functions in the KL expansion as random functions. We think of the first set of uncertainties as associated with a coarse scale model, and of the second set as associated with subscale fluctuations not captured in the coarse scale description.