Model reduction in computational homogenization for transient heat conduction (Q2665032)

The authors propose a computationally efficient homogenization method for transient heat conduction problems posed in non-homogeneous media. They consider the transient heat problem (energy balance) \(\overline{\nabla } \cdot \overline{q}+\overline{\overset{.}{\varepsilon }}=0\) where \(\overline{q }\) and \(\overline{\overset{.}{\varepsilon }}\) are the macroscopic heat flow and the rate of change of macroscopic internal energy.\ At a microscale, the problem is written as \(\nabla \cdot q+\overset{.}{\varepsilon }=0\) and Fourier's law \(q=-\lambda \nabla \theta \) is added. These equations may be completed with boundary conditions. The authors then present the downscaling and upscaling process to transfer information concerning the temperature between the two scales. They introduce a model order reduction using a finite element formulation, first considering tied and retained nodes, writing a microscale discrete problem and decomposing the microscopic solution in a steady-state and a transient components. They here assume a linearity hypothesis and a relaxed separation of scales. This allows to derive the macroscopic quantities through an averaging process. In the last part of their paper, the authors apply their method in the case of a microstructure containing randomly distributed circular inclusions. They present the steady-state and transient reduced bases and the results they obtain together with the computational cost. The paper ends with a presentation of the applicability limits of this method in the case where significant transient phenomena occur.