Uncertainty propagation for effective reduced-order model generation

This work describes a procedure to quantify effective uncertainty propagation through the development of a reduced-order model. To accomplish this objective, the concept of a random fuzzy variable is applied to represent both random and systematic errors associated with uncertain variables. A procedure to obtain feasible combinations of multiple uncertain variables is described. To predict the output probabilistic measure accurately with a minimum number of sample, efficient sampling that combines the techniques of Latin hypercube sampling and Hammersley sequence pairing is used. Based on the output data a reduced-order model is generated using the well known Karhunen-Loève expansion. The results show that the outputs of the reduced-order model track the outputs of the nonlinear physics-based model satisfactorily. A chemical reactor is used to demonstrate the concepts.