Deep learning for automated characterization of pore-scale wettability

https://doi.org/10.1016/j.advwatres.2020.103708Get rights and content

Abstract

A procedure combining experiments and deep learning is demonstrated to acquire pore-scale images of oil- and water-wet surfaces over a large field of view in microfluidic devices and to classify wettability based upon these pore scale images. Deep learning supplants the manual, time-consuming, error-prone investigation and categorization of such images. Image datasets were obtained by visualizing the distribution of immiscible phases (n-decane and water) within in-house fabricated micromodels containing sandstone-type and carbonate-type pore structures. The reference dataset consists of 6400 color images binned into four classes for sandstone (water- or oil-wet surfaces) and carbonate (water- or oil-wet surfaces) pore-network patterns. There are 1600 images per class. During 10 sequential training and testing runs of the deep-learning algorithm, 3000, 100, and 100 images were randomly assigned per each rock pattern as the training, validation, and test sets, respectively. We trained and optimized both a Fully Connected Neural Network (FCN) and Convolutional Neural Network (ConvNet) using the image data. The ConvNet performs better as 5 and 8 layers are implemented, as expected. The FCN shows an average test set accuracy for binary surface wettability classification of 87.4% for sandstone rock type and 98.7% for carbonate rock type pore networks. Distinctive heterogeneity in the carbonate rock type and its relevant phase saturation profile resulted in a better prediction accuracy. The best ConvNet models shows an average test set accuracy of binary surface wettability classification of 99.4 ± 0.1% for both sandstone-type and carbonate pore networks. Heterogenous pore sizes and an abundance of small pores amplify the effects of wetting and aid identification. Overall, the test set accuracy for the simultaneous classification of four classes including both sandstone (water- or oil-wet) and carbonate rock pattern (water- or oil-wet) is 98.5% with an 8-layer ConvNet. Performance of the deep-learning model is further interpreted using saliency maps that indicate the degree to which each pixel in the image affects the classification score. Pixels at and adjacent to interfaces are most important to classification.

Introduction

Coupled physical mechanisms resulting from viscous, capillary, and surface forces make the description of multiphase flow in porous media challenging. At the pore scale, the interplay of pore network geometry, connectivity, and surface wettability play an important role in the distribution of fluid phases. Improved understanding of subsurface fluid flows, and specifically the pore scale, is required to facilitate critical processes in the context of water and energy resources, including recovery of spilled hydrocarbons, enhanced oil recovery, and geological carbon sequestration.

Critical understanding of multiphase flow has been enabled by microfluidic devices known as micromodels, advanced imaging technologies, and improved computational power (Armstrong et al., 2016). Micromodels contain microlevel representations of rock structure and are often made of silicon, e.g., Buchgraber et al. (2012a), glass. or HDMS. The strength of microfluidics for characterization of multiphase flow effects in porous media include reproducibility and control over pore structures, optical access to pore-scale events, high resolution images, and pore-scale controls of fluid flow. In the context of subsurface water and energy resources, micromodels reveal mechanisms controlling multiphase flow and transport phenomena in the pore space of geological porous media (Sinton, 2014). For instance, understanding fluid flow through dual-porosity carbonate rock is complex due to its heterogeneous pore network across multiple length scales. A typical dual-porosity carbonate pore network consists of a wide size range of pores (1–200 µm) and minerals mixed with micro- and macroscale fractures and particles of varying size (Yun et al., 2017).

Micromodels are often criticized for their lack of reproduction of solid/fluid chemistry whereas the effect of wettability (i.e., the affinity of fluid for a pore wall) on fluid movement can be studied and visualized particularly well in the pore space of micromodels (Zhao et al., 2016). A typical etched-silicon micromodel has uniform silicon dioxide pore surfaces after manufacture. Accordingly, recent papers have emphasized the functionalization of micromodels with clay minerals and calcite (Song, Ogunbanwo, Steinsbø, Fernø, Kovscek, 2018, Song, Kovscek, 2015, Yun, Chang, Cogswell, Eichmann, Gizzatov, Thomas, Al-Hazza, Abdel-Fattah, Wang, 2020) as well as construction of micromodels out of calcite (Song et al., 2014). Other studies have focused on the understanding of how chemicals alter the arrangement of fluids at pore scale (Howe, Clarke, Mitchell, Staniland, Hawkes, Whalan, 2015, Yun, Chang, Cogswell, Eichmann, Gizzatov, Thomas, Al-Hazza, Abdel-Fattah, Wang, 2020).

The accurate and rapid characterization of trapping processes, the arrangement of fluids at pore scale, and the impact of trapping on multiphase displacement in micromodels remains a challenge. In short, it is not yet currently feasible, to perform quick and thorough interrogation of a large number of sample images of rock and its pore space at high resolution over a large Field of View (FOV). Thus, full characterization of the interactions between aqueous phases, organic phases, and rock is limited. We believe that pore-to-Darcy scale observation of displacement processes in complex and realistic porous media (e.g., sandstone-type and carbonate-type systems) can be achieved by fast monitoring using a large number of microscopic images of pores existing in porous media.

Motivated by advances in related fields that use microfluidics (Riordon et al., 2019), we explore machine learning for the rapid and accurate categorization of pore-scale images. We place special emphasis on the fluids that wet pore walls because of the importance of wettability to multiphase flow (Blunt et al., 2013) and the various industrially-relevant methods employed to change wettability at the pore scale (Lu, Goudarzi, Chen, Kim, Delshad, Mohanty, Sepehrnoori, Weerasooriya, Pope, 2014, Hirasaki, Miller, Puerto, 2008) Image analysis to quantify the volume fraction of pore space filled by aqueous, organic, and solid phases is relatively easy to automate, but the interpretation of the fluids that wet solid surfaces remains painstaking and laborious. Hence, the main objective of this manuscript is to explore the application of deep learning to enable automatic categorization of wettability using pore-scale images. We contribute here to understanding of multiphase flow through aiding the experimentalist. Toward this purpose, this paper proceeds with a literature review including wettability, microfluidics for multiphase flow applications, and applications of machine learning to porous media images. Then the methodologies for image acquisition and the training of deep learning models to classify wettability using Fully Connected Layer Networks (FCN) and Convolutional Neural Networks (ConvNet) are discussed. Image classification results follow including a saliency analysis that illustrates qualitatively and quantitatively the role of each pixel on image classification.

Section snippets

Literature review

This section discusses the importance of understanding the impact of wettability of the porous system where multiple phases (oil, water, and gas) coexist. Micromodels as a means of understanding wettability and flow mechanisms are also discussed. Then, recent studies employing machine learning for accelerating data analysis are presented.

Data acquisition

An experimental procedure to fabricate microfluidic devices with strong oil- and water-wet surface wetting conditions is presented. Importantly, Laser Scanning Fluoresence Microscopy (LSFM) is used to collect many pore-scale images. Image processing, post image acquisition, and details of the data structure for images are also presented in this section to document the use of microfluidic image data and deep-learning.

Training

Spatial resolution is important for wettability classification. Hence, we explain the selection criteria for the 6400 images with different FOVs for all classes. This section discusses our methodology using fully connected neural networks and convolutional neural network models in detail. Its also details the evaluation criteria.

Results and discussion

The wettability and morphology of pores strongly influences the ganglia shape, and pore-scale configuration of oil and water phases. This section presents the predictions of wettability (strong oil-wetting or strong water-wetting) and illustrates the interplay between the oil phase, water phase, and grain surfaces.

Summary and future work

This work presented and evaluated a methodology for automated classification of images from microfluidic devices (micromodels) with water- and oil-wet surface wettability. Fluid injection experiments using micromodels with two different pore network patterns were conducted to prepare pore-level images that were used as a data-set to train two deep-learning algorithms. The first is a three-layer fully connected neural network (FCN) and the second is a convolutional neural network (ConvNet). Both

CRediT authorship contribution statement

Wonjin Yun: Conceptualization, Methodology, Software, Writing - original draft. Yimin Liu: Conceptualization, Methodology, Software, Writing - original draft. Anthony R. Kovscek: Conceptualization, Supervision, Funding acquisition, Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgment

The support of the SUPRI-A Industrial Affiliates is acknowledged gratefully.

References (40)

  • A.M. Alhammadi et al.

    In situ characterization of mixed-wettability in a reservoir rock at subsurface conditions

    Sci. Rep.

    (2017)
  • N. Alqahtani et al.

    Deep Learning Convolutional Neural Networks to Predict Porous Media Properties

    SPE Asia Pacific Oil and Gas Conference and Exhibition

    (2018)
  • R.T. Armstrong et al.

    Beyond Darcy’s law: the role of phase topology and ganglion dynamics for two-fluid flow

    Phys. Rev. E

    (2016)
  • J. Avendaño et al.

    Effect of surface wettability on immiscible displacement in a microfluidic porous media

    Energies

    (2019)
  • M. Buchgraber et al.

    A study of microscale gas trapping using etched silicon micromodels

    Transp. Porous Media

    (2012)
  • I. Goodfellow et al.

    Deep Learning

    (2016)
  • J.W. Grate et al.

    Silane modification of glass and silica surfaces to obtain equally oil-wet surfaces in glass-covered silicon micromodel applications

    Water Resour. Res.

    (2013)
  • K. He et al.

    Deep residual learning for image recognition

    Proceedings of the IEEE conference on computer vision and pattern recognition

    (2016)
  • G.J. Hirasaki et al.

    Recent advances in surfactant eor

    SPE Annual Technical Conference and Exhibition

    (2008)
  • R. Holtzman

    Effects of pore-Scale disorder on fluid displacement in partially-Wettable porous media

    Sci. Rep.

    (2016)
  • Cited by (11)

    View all citing articles on Scopus
    View full text