The problem of model order reduction in the analysis of large complicated numericalmodels is considered. The objective of the model order reduction is to minimise the computational costs without loss of accuracy. In this study, the focus is on linear dynamical systems with parametric model uncertainties. A twin approach to model order reduction is adopted. Efficient treatment of the model uncertainties is carried out by projecting the uncertain parameters into the Hilbert subspace. This enables expressing the random eigenvalue problem in terms of polynomial chaos expansions (PCE) whose dimensions are restricted to only the first few dominant stochastic modes. Reduction in the sizes of the global matrices used in the random eigenvalue analysis is obtained through system equivalent reduction expansion process (SEREP) which retains only the dominant dynamic modes in the analysis. Preliminary numerical results are presented to demonstrate the proposed PCE based SEREP as a model reduction strategy.

Keywords: Reduced order modeling, SEREP, PCE, Random eigenvalue problem.