Journal
ROCK MECHANICS AND ROCK ENGINEERING
Volume 55, Issue 1, Pages 235-248Publisher
SPRINGER WIEN
DOI: 10.1007/s00603-021-02668-9
Keywords
ROM; Numerical modelling; Recurrent neural networks
Funding
- AGL Loy Yang
Ask authors/readers for more resources
Fully coupled hydro-mechanical simulations of fractured media require sophisticated non-linear solvers to capture the complex relationship between fluid flow and material's mechanical response. Modelling these systems can be onerous, so a reduction strategy is necessary to predict physical response with less computational effort and time.
Fully coupled hydro-mechanical simulations of fractured media require sophisticated non-linear solvers to capture the complex relationship between fluid flow and a material's mechanical response. Such simulations may involve detailed meshes comprising millions of degrees of freedom. As a result, modelling these systems can be quite onerous, with the number of runs usually limited to a few realizations for practical purposes. Nevertheless, in many cases, it may be necessary to understand a system's behaviour over long timescales or subject to a broad range of boundary conditions. In such cases, a reduction strategy is necessary to obtain models that are able to predict the relevant physical response for significantly less computational effort and time. In this paper, we describe a strategy to develop reduced order models (or ROMs) from fully coupled hydro-mechanical simulations of fractured media. We outline the key steps involved in the development of the ROMs, and highlight some of the potential challenges. To illustrate the approach, we derive an ROM for a rectilinear region subject to simple loading conditions. While the high-fidelity simulations used to create the ROM require several days of central processing unit (CPU) time on a dedicated cluster, the new reduced order model is able to conduct scenario assessments involving a thousand simulations in a matter of minutes on a single CPU.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available