Multiscale modeling of prismatic heterogeneous structures based on a localized hyperreduced-order method

This work aims at deriving special types of one-dimensional Finite Elements (1D FE) for efficiently modeling heterogeneous prismatic structures, in the small strains regime, by means of reduced-order modeling (ROM) and domain decomposition techniques. The employed partitioning framework introduces fictitious interfaces between contiguous subdomains, leading to a formulation with both subdomain and interface fields. We propose a low-dimensional parameterization at both subdomain and interface levels by using reduced-order bases precomputed in an offline stage by applying the Singular Value Decomposition (SVD) on solution snapshots. In this parameterization, the amplitude of the fictitious interfaces play the role of coarse-scale displacement unknowns. We demonstrate that, with this partitioned framework, it is possible to arrive at a solution strategy that avoids solving the typical nested local/global problem of other similar methods (such as the FE2 method). Rather, in our approach, the coarse-grid cells can be regarded as special types of finite elements, whose nodes coincides with the centroids of the interfaces, and whose kinematics are dictated by the modes of the fictitious interfaces. This means that the kinematics of our coarse-scale FE are not pre-defined by the user, but extracted from the set of training computational experiments. Likewise, we demonstrate that the coarse-scale and fine-scale displacements are related by inter-scale operators that can be precomputed in the offline stage. Lastly, a hyperreduced scheme is considered for the evaluation of the internal forces, allowing us to deal with possible material nonlinearities. (c) 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Automated summary

This study aims to efficiently model heterogeneous prismatic structures under small strains regime using reduced-order modeling (ROM) and domain decomposition techniques. By introducing fictitious interfaces and a low-dimensional parameterization scheme, the proposed partitioning framework avoids solving the nested local/global problem of other methods. The kinematics of the coarse-scale finite elements are not predefined by the user, but extracted from training computational experiments.