Deep learning–driven permeability estimation from 2D images

Current micro-CT image resolution is limited to 1–2 microns. A recent study has identified that at least 10 image voxels are needed to resolve pore throats, which limits the applicability of direct simulations using the digital rock (DR) technology to medium-to-coarse–grained rocks (i.e., rocks with permeability > 100 mD). On the other hand, 2D high-resolution colored images such as the ones obtained from transmitted light microscopy delivers a much higher resolution (approximately 0.6 microns). However, reliable and efficient workflows to jointly utilize full-size 2D images, measured 3D core-plug permeabilities, and 2D direct pore-scale ow simulations on 2D images within a predictive framework for permeability estimation are lacking. In order to close this gap, we have developed a state-of-the-art deep learning (DL) algorithm for the direct prediction of permeability from 2D images. We take advantage of the computing graphics processing units (GPUs) in our implementation of this algorithm. The trained DL model predicts properties accurately within seconds, and therefore, provide a significant speeding up simulation workflow. A real-life dataset is used to demonstrate the applicability and versatility of the proposed method.

[1]  Michael Zibulevsky,et al.  Block-based compressed sensing of images via deep learning , 2017, 2017 IEEE 19th International Workshop on Multimedia Signal Processing (MMSP).

[2]  Gary Mavko,et al.  Estimating permeability from thin sections without reconstruction: Digital rock study of 3D properties from 2D images , 2017, Comput. Geosci..

[3]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[4]  Nattavadee Srisutthiyakorn,et al.  Deep-learning methods for predicting permeability from 2D/3D binary-segmented images , 2016 .

[5]  Guigang Zhang,et al.  Deep Learning , 2016, Int. J. Semantic Comput..

[6]  Mathias Unberath,et al.  Deep Learning Computed Tomography: Learning Projection-Domain Weights From Image Domain in Limited Angle Problems , 2018, IEEE Transactions on Medical Imaging.

[7]  Geoffrey E. Hinton,et al.  On the importance of initialization and momentum in deep learning , 2013, ICML.

[8]  Ge Wang,et al.  A Perspective on Deep Imaging , 2016, IEEE Access.

[9]  Sebastian Ruder,et al.  An overview of gradient descent optimization algorithms , 2016, Vestnik komp'iuternykh i informatsionnykh tekhnologii.

[10]  Geoffrey E. Hinton,et al.  Deep Learning , 2015, Nature.

[11]  Taha Sochi,et al.  Pore-Scale Modeling of Non-Newtonian Flow in Porous Media , 2010, 1011.0760.

[12]  F. Alpak,et al.  Imaging and computational considerations for image computed permeability: Operating envelope of Digital Rock Physics , 2018, Advances in Water Resources.

[13]  Thomas Brox,et al.  U-Net: Convolutional Networks for Biomedical Image Segmentation , 2015, MICCAI.

[14]  Taha Sochi,et al.  Non-Newtonian flow in porous media , 2010 .

[15]  Amir Adler,et al.  Deep-learning tomography , 2018 .

[16]  John Tran,et al.  cuDNN: Efficient Primitives for Deep Learning , 2014, ArXiv.

[17]  Yoshua Bengio,et al.  Algorithms for Hyper-Parameter Optimization , 2011, NIPS.

[18]  Evgeny Burnaev,et al.  Driving Digital Rock towards Machine Learning: predicting permeability with Gradient Boosting and Deep Neural Networks , 2018, Comput. Geosci..

[19]  Yuan Yu,et al.  TensorFlow: A system for large-scale machine learning , 2016, OSDI.