Guaranteed error bounds in homogenisation: an optimum stochastic approach to preserve the numerical separation of scales

Abstract

This paper proposes a new methodology to guarantee the accuracy of the homogenisation schemes that are traditionally employed to approximate the solution of PDEs with random, fast evolving diffusion coefficients. More precisely, in the context of linear elliptic diffusion problems in randomly packed particulate composites, we develop an approach to strictly bound the error in the expectation and second moment of quantities of interest, without ever solving the fine‐scale, intractable stochastic problem. The most attractive feature of our approach is that the error bounds are computed without any integration of the fine‐scale features. Our computations are purely macroscopic, deterministic and remain tractable even for small scale ratios. The second contribution of the paper is an alternative derivation of modelling error bounds through the Prager–Synge hypercircle theorem. We show that this approach allows us to fully characterise and optimally tighten the interval in which predicted quantities of interest are guaranteed to lie. We interpret our optimum result as an extension of Reuss–Voigt approaches, which are classically used to estimate the homogenised diffusion coefficients of composites, to the estimation of macroscopic engineering quantities of interest. Finally, we make use of these derivations to obtain an efficient procedure for multiscale model verification and adaptation. Copyright © 2016 John Wiley & Sons, Ltd.

A Bayesian multiscale CNN framework to predict local stress fields in structures with microscale features

“…When the scales cannot be separated, scientists resort to domain decomposition-based approach. The results of homogenisation are applied to the boundary of regions of interest for concurrent microscale corrections to be performed [15,24,39,40,42]. These approaches are computationally more expensive and practically more intrusive than methods based on RVEs.…”

A posteriori error estimation and adaptive strategy for the control of MsFEM computations

“…On the one hand, a large literature has addressed a priori error estimation for the last fifteen years, enabling to evaluate the convergence of the error with respect to the macroscopic mesh size and the characteristic length of the microscopic heterogeneities [6,25,26,27,7,28]. On the other hand, despite the fact that a posteriori error control and adaptivity have become an important issue for reliable and efficient multiscale computations, very few tools on this topic are available (see [29,30,31,32,33,34,35,36]) and there are many open questions: quantitative assessment of error propagation across scales, relevant adaptive strategies, . .…”

Random Stiffness Tensor of Particulate Composites with Hyper-Elastic Matrix and Imperfect Interface

“…In such cases, the representative volume element is selected to perfectly reflect the complex material microstructure and, thanks to the application of the homogenization method, to determine its effective properties whilst using a relatively small computational effort. Theoretical and numerical error bounds for macroscopic material characteristics can be estimated even before the final homogenization [15], but they may be very wide, especially for a larger contrast of the constituents' properties. Composites after homogenization are assumed to be elastic and isotropic solids, but such an approach is perfect for small strain engineering applications and returns remarkable modelling errors in the case of incremental or dynamic loading and for the irregular microstructures.…”