Quantification of Polymorphic Uncertainties: A Quasi-Monte Carlo Approach

Abstract

A novel methodology for the quantification of mixed and nested polymorphic uncertainties has been developed. It is designed to be applied to moderately computationally expensive deterministic models, and it preserves the distinction between aleatory and epistemic uncertainties throughout the entire process. Quasi-Monte Carlo sampling is used to efficiently represent the local and global behavior of the models. By decoupling uncertainty propagation and processing, the methodology achieves efficient reuse of samples and can support multiple outputs. In addition, simultaneous estimation of sensitivity indices is possible to facilitate decisions on where to reduce epistemic uncertainties. It is demonstrated on a structural dynamics example and compared with a fully stochastic approach using the pignistic transform. The proposed methodology has been demonstrated to be significantly more efficient than a naive implementation, but adds computational cost compared with a fully stochastic approach.