Bayesian inversion for steady flow in fractured porous media with contact on fractures and hydro mechanical coupling

0533163 - ÚGN 2021 RIV NL eng J - Journal Article
Blaheta, Radim - Béreš, Michal - Domesová, Simona - Horák, David
Bayesian inversion for steady flow in fractured porous media with contact on fractures and hydro mechanical coupling.
Computational Geosciences. Roč. 24, February 2020 (2020), s. 1911-1932. ISSN 1420-0597. E-ISSN 1573-1499
R&D Projects: GA ČR(CZ) GA19-11441S; GA MŠMT ED1.1.00/02.0070; GA MŠMT LQ1602
Institutional support: RVO:68145535
Keywords : porous media with fractures * coupled hydro-mechanics * Bayesian inversion * multi-dimensional flow model * contact mechanics on fractures
OECD category: Applied mathematics
Impact factor: 2.413, year: 2020
Method of publishing: Limited access
https://link.springer.com/article/10.1007/s10596-020-09935-8

The paper is motivated by a strong interest in numerical analysis of flow in fractured porous media, e.g., rocks in geo-engineering applications. It follows the conception of porous media as a continuum with fractures which are represented as lower dimensional objects. In the paper, the finite element discretization of the flow in coupled continuum and fractures is used. Fluid pressures serve as the basic unknowns. In many applications, the flow is connected with deformations of the porous matrix, therefore, the hydro-mechanical coupling is also considered. The fluid pressure is transferred to the mechanical load in both pores and fractures and the considered mechanical model involves elastic deformations of the porous matrix and opening/closing of the fractures with the non-penetration constraint. The mechanical model with this constraint is implemented via the technique of the Lagrange multipliers, duality formulation, and combination with a suitable domain decomposition method. There is usually lack of information about problem parameters and they undergo many uncertainties coming e.g. from the heterogeneity of rock formations and complicated realization of experiments for parameter identification. These experiments rarely provide some of the asked parameters directly but require solving inverse problems. The stochastic (Bayesian) inversion is natural due to the mentioned uncertainties. In this paper, the implementation of the Bayesian inversion is realized via Metropolis-Hastings Markov chain Monte Carlo approach. For the reduction of computational demands, the sampling procedure uses the delayed acceptance of samples based on a surrogate model which is constructed during a preliminary sampling process. The developed hydro-mechanical model and the implemented Bayesian inversion are tested on two types of model inverse problems.
Permanent Link: http://hdl.handle.net/11104/0311636