Exploiting POD-based model order reduction for damped structural dynamics without damped training data

Resumo

The Proper Orthogonal Decomposition (POD) model order reduction technique can identify underlying patterns and reproduce system configurations beyond those present in the training dataset. In this study, we investigate this capability through the simulation of proportional Rayleigh damping matrices, in structural dynamic problems, despite the absence of damping samples in the algorithm's training phase. This aspect of the method remains underexplored, despite its potential to significantly reduce computational effort by avoiding the simulation of the damping matrices in the physical domain (in addition to the computational savings already afforded by the reduced order itself). The proposed strategy is applied to the solution of structural problems discretized via the Finite Element Method (FEM) and time integration methods based on numerical Green’s functions (namely, ImGA and ExGA). The results are then compared with those obtained through standard solution approaches.