Frequency domain analysis of model order reduction techniques

Abstract

In this Study, some popular model order reduction and superelement techniques are studied in frequency and time domains. Frequency domain identification (FDI) methods are also applied to semidiscrete finite element equations to obtain reduced order models for structural systems. An FDI algorithm that determines the coefficients of the reduced order model by using nonlinear least squares (NLS) method and subspace based identification (SBI) method are applied to a sample problern and the results are compared with component mode synthesis (CMS) and quasi-static mode synthesis (QSM) methods. In literature, phase errors of model order reduction techniques have not been studied; however, it is shown in this paper that phase errors are also important in evaluating the performance of model order reduction techniques. In general, the NLS and SBI methods have better performance than the CMS and QSM methods; however, as the size of problems increases, the NLS method may have convergence problems and the SBI method may yield estimated models having large orders