TY - GEN

T1 - Numerical Experiments with Digital Twins of Core Samples for Estimating Effective Elastic Parameters

AU - Reshetova, Galina

AU - Cheverda, Vladimir

AU - Khachkova, Tatyana

PY - 2019

Y1 - 2019

N2 - The use of 3D Computed Tomography (CT) throws light on the interior structure of the core samples. That is why the 3D CT has become increasingly widespread in digital rock physics. However, the CT does not provide information about the elastic properties of the rocks. So far, the most common approach to recover these properties is static laboratory measurements, when a sample is subject to a variety of static loads, followed by the measurement of displacements. This way is rather expensive, time-consuming and sometimes result in the destruction of a sample. Instead, we propose to implement numerical experiments by solving a 3D elastic static problem to perform upscaling of the CT data and to compute effective elastic parameters. In fact, to find these parameters we need to resolve a huge system of linear algebraic equations, which is a troublesome task. We have decided not to use a direct solver, but to apply a new iterative relaxation technique by considering a dynamic elastic problem with a special choice of relaxation parameters. The approach proposed needs parallelization. We have come to the conclusion to use the new feature of the latest Fortran extension known as CoArray Fortran (CAF) and to compare the three ways of parallel implementation as applied to this problem: MPI, MPI+OpenMP and CAF. Our experiments have proved that CAF and MPI approximately demonstrate the same performance, but the CAF possesses a clearer structure and is easy for coding.

AB - The use of 3D Computed Tomography (CT) throws light on the interior structure of the core samples. That is why the 3D CT has become increasingly widespread in digital rock physics. However, the CT does not provide information about the elastic properties of the rocks. So far, the most common approach to recover these properties is static laboratory measurements, when a sample is subject to a variety of static loads, followed by the measurement of displacements. This way is rather expensive, time-consuming and sometimes result in the destruction of a sample. Instead, we propose to implement numerical experiments by solving a 3D elastic static problem to perform upscaling of the CT data and to compute effective elastic parameters. In fact, to find these parameters we need to resolve a huge system of linear algebraic equations, which is a troublesome task. We have decided not to use a direct solver, but to apply a new iterative relaxation technique by considering a dynamic elastic problem with a special choice of relaxation parameters. The approach proposed needs parallelization. We have come to the conclusion to use the new feature of the latest Fortran extension known as CoArray Fortran (CAF) and to compare the three ways of parallel implementation as applied to this problem: MPI, MPI+OpenMP and CAF. Our experiments have proved that CAF and MPI approximately demonstrate the same performance, but the CAF possesses a clearer structure and is easy for coding.

KW - 3d tomographic images

KW - Coarray fortran

KW - Digital core

KW - Elastic static stresses

KW - MPI

KW - Virtual experiments

UR - http://www.scopus.com/inward/record.url?scp=85076862113&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-36592-9_24

DO - 10.1007/978-3-030-36592-9_24

M3 - Conference contribution

AN - SCOPUS:85076862113

SN - 9783030365912

T3 - Communications in Computer and Information Science

SP - 290

EP - 301

BT - Supercomputing - 5th Russian Supercomputing Days, RuSCDays 2019, Revised Selected Papers

A2 - Voevodin, Vladimir

A2 - Sobolev, Sergey

PB - Springer Netherlands

T2 - 5th Russian Supercomputing Days Conference, RuSCDays 2019

Y2 - 23 September 2019 through 24 September 2019

ER -