Physics-informed neural network simulation of multiphase poroelasticity using stress-split sequential training

Haghighat, Ehsan; Amini, Danial; Juanes, Rubén

doi:10.1016/j.cma.2022.115141

Cited by 76 publications

(19 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…x, mm and non-dimensional Euler equations in training and found that the former physics loss is far less stable. 8 This finding confirms the numerical tests of Haghighat et al [83], who attributed the performance differential to the disparate magnitudes of the dimensional Euler loss components, spanning roughly five orders of magnitude in our scenarios. An adaptive weighting scheme could potentially help to overcome this issue, but we previously observed that such schemes are unstable for noisy data [32].…”

Section: Planar Expansion Fansupporting

confidence: 89%

Estimating density, velocity, and pressure fields in supersonic flow using physics-informed BOS

Molnar¹,

Venkatakrishnan²,

Schmidt³

et al. 2022

Preprint

View full text Add to dashboard Cite

We report a new workflow for background-oriented schlieren (BOS), termed "physics-informed BOS," to extract density, velocity, and pressure fields from a pair of reference and distorted images. Our method uses a physics-informed neural network (PINN) to produce flow fields that simultaneously satisfy the measurement data and governing equations. For the high-speed flows of interest in this work, we specify a physics loss based on the Euler and irrotationality equations. BOS is a quantitative fluid visualization technique that is used to characterize high-speed flows. Images of a background pattern, positioned behind the target flow, are processed using computer vision and tomography algorithms to determine the density field. Crucially, BOS features a series of ill-posed inverse problems that require supplemental information (i.e., in addition to the images) to accurately reconstruct the flow. Current BOS workflows rely upon interpolation of the images or a penalty term to promote a globally-or piecewise-smooth solution. However, these algorithms are invariably incompatible with the flow physics, leading to errors in the density field. Physics-informed BOS directly reconstructs all the flow fields using a PINN that includes the BOS measurement model and governing equations. This procedure improves the accuracy of density estimates and also yields velocity and pressure data, which was not previously available. We demonstrate our approach by reconstructing synthetic data that corresponds to analytical and numerical phantoms as well as experimental measurements. Our physics-informed reconstructions are significantly more accurate than conventional BOS estimates. Further, to the best of our knowledge, this work represents the first use of a PINN to reconstruct a supersonic flow from experimental data of any kind.

show abstract

Section: Planar Expansion Fansupporting

confidence: 89%

Estimating density, velocity, and pressure fields in supersonic flow using physics-informed BOS

Molnar¹,

Venkatakrishnan²,

Schmidt³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…have not yet shown significant advantages in computational efficiency for subsurface multiphase flow through heterogeneous media [44][45][46][47] . On the contrary, as a datadriven approach, Nested FNO provides a framework with great potential not only for CO 2 storage but also for other scientific problems that involve multi-level modeling.…”

Section: Discussionmentioning

confidence: 99%

Accelerating Carbon Capture and Storage Modeling using Fourier Neural Operators

Wen¹,

Li²,

Long³

et al. 2022

Preprint

View full text Add to dashboard Cite

Carbon capture and storage (CCS) is an important strategy for reducing carbon dioxide emissions and mitigating climate change. We consider the storage aspect of CCS, which involves injecting carbon dioxide into underground reservoirs. This requires accurate and high-resolution predictions of carbon dioxide plume migration and reservoir pressure buildup. However, such modeling is challenging at scale due to the high computational costs of existing numerical methods. We introduce a novel machine learning approach for four-dimensional spatial-temporal modeling, which speeds up predictions nearly 700,000 times compared to existing methods. It provides highly accurate predictions under diverse reservoir conditions, geological heterogeneity, and injection schemes. Our framework, Nested Fourier Neural Operator (FNO), learns the solution operator for the family of partial differential equations governing the carbon dioxide-water multiphase flow. It uses a hierarchy of FNO models to produce outputs at different refinement levels. Thus, our approach enables unprecedented real-time high-resolution modeling for carbon dioxide storage.

show abstract

“…Haghighat et al (Haghighat et al, 2021) applied PINN to solve the equations of coupled flow and deformation of singlephase and multi-phase flows in fluid flow problems in porous media. They chose the dimensionless form of the coupling relationship of porous media under single-phase or two-phase flow conditions in the loss function, thus avoiding the unstable optimization problem that may be caused by multiple differential relationships.…”

Section: Methods Used Referencementioning

confidence: 99%

“…Previous studies (Papaliangas et al, 1993;Sklavounos and Sakellariou, 1995;Indraratna et al, 1999;Wang et al, 2000;Jin et al, 2006;Vardakos. et al, 2007;Tiryaki, 2008;Zorlu et al, 2008;Ceryan et al, 2012;Samani and Bafghi, 2012;Yagiz et al, 2012;Rajesh Kumar et al, 2013;Zheng et al, 2013;Gu et al, 2015;Liu et al, 2015;Song et al, 2015;Wang et al, 2015;Raissi et al, 2018;Matos et al, 2019;Almajid and Abu-Alsaud, 2021;Dantas Neto et al, 2021;Deng and Pan, 2021;Haghighat et al, 2021;Hasanipanah et al, 2021;Fathipour-Azar, 2022;Garg et al, 2022;Li and Chen, 2022;Mahmoodzadeh et al, 2022) have shown that machine learning methods can provide new ideas for rock mechanics problems, and this issue is no exception (Ghaboussi and Sidarta, 1998;Jaksa. and Maier., 2009).…”

Section: Determining Constitutive Behaviorsmentioning

confidence: 99%

Machine learning for rock mechanics problems; an insight

Taleghani

Balushi

et al. 2022

Front. Mech. Eng.

View full text Add to dashboard Cite

Due to inherent heterogeneity of geomaterials, rock mechanics involved with extensive lab experiments and empirical correlations that often lack enough accuracy needed for many engineering problems. Machine learning has several characters that makes it an attractive choice to reduce number of required experiments or develop more effective correlations. The timeliness of this effort is supported by several recent technological advances. Machine learning, data analytics, and data management have expanded rapidly in many commercial sectors, providing an array of resources that can be leveraged for subsurface applications. In the last 15 years, deep learning in the form of deep neural networks, has been used very effectively in diverse applications, such as computer vision, seismic inversion, and natural language processing. Despite the remarkable success in these and related areas, deep learning has not yet been widely used in the field of scientific computing specially when it comes to subsurface applications due to the lack of large amount of data to train algorithms. In this paper, we review such efforts and try to envision future game-changing advances that may impact this field.

show abstract

Physics-informed neural network simulation of multiphase poroelasticity using stress-split sequential training

Cited by 76 publications

References 51 publications

Estimating density, velocity, and pressure fields in supersonic flow using physics-informed BOS

Estimating density, velocity, and pressure fields in supersonic flow using physics-informed BOS

Accelerating Carbon Capture and Storage Modeling using Fourier Neural Operators

Machine learning for rock mechanics problems; an insight

Contact Info

Product

Resources

About