Surrogate models for transient phenomena in poromechanics

Tailing dams constitute critical geotechnical infrastructure designed for the containment of mining waste. Its stability is governed by the complex coupling between pore fluid pressures and soil deformation [1]. Accurately modelling this behavior requires coupled porous hydro-mechanical simulations; however, the computational cost associated with high-fidelity finite element models is often unaffordable for extensive sensitivity analyses or real-time monitoring.

To overcome this bottleneck, dimensionality reduction techniques offer a robust and efficient alternative. The application of these techniques involves two phases: i) creation of a training set via a sampling of the parameter space, and ii) creation of a reduced space where to find new solutions within the family.

In this work, we explore the use of Singular Value Decomposition (SVD) and, specifically, High Order Singular Value Decomposition (HOSVD), to build multidimensional surrogates of the transient-hydromechanical models capable of accurately reproducing solutions within a fraction of the computational time.

 

REFERENCES

• Nasika C., P. Díez, P. Gerard, T.J. Massart and S. Zlotnik. Towards real time assessment of earthfill dams via Model Order Reduction. Finite Elements in Analysis & Design, Vol. 199, 103666, doi:10.1016/j.finel.2021.103666, 2022.