Reduced order models for geometrically nonlinear structures: Assessment of implicit condensation in comparison with invariant manifold approach

A comparison between two methods to derive reduced-order models (ROM) for geometrically nonlinear structures is proposed. The implicit condensation and expansion (ICE) method relies on a series of applied static loadings. From this set, a stress manifold is constructed for building the ROM. On the other hand, nonlinear normal modes rely on invariant manifold theory in order to keep the key property of invariance for the reduced subspaces. When the model coefficients are fully known, the ICE method reduces to a static condensation. However, in the framework of finite element discretization, getting all these coefficients is generally too computationally expensive. The stress manifold is shown to tend to the invariant manifold only when a slow/fast decomposition between master and slave coordinates can be assumed. Another key problem in using the ICE method is related to the fitting procedure when a large number of modes need to be taken into account. A simplified procedure, relying on normal form theory and identification of only resonant monomial terms in the nonlinear stiffness, is proposed and contrasted with the current method. All the findings are illustrated on beams and plates examples.

Automated summary

This study compares two methods for deriving reduced-order models for geometrically nonlinear structures: the ICE method and nonlinear normal modes. The ICE method relies on static loadings to construct a stress manifold, while nonlinear normal modes depend on invariant manifold theory to maintain invariance. In the finite element framework, obtaining all model coefficients for the ICE method is often computationally expensive, and the stress manifold tends to the invariant manifold in specific conditions.