A novel deep learning-assisted reservoir fracture delineation with Conditional Generative Adversarial Networks

Ardiansyah Koeshidayatullah

Research output: Chapter in Book/Report/Conference proceedingConference contribution


Fracture characterization and delineation have emerged as one of the key parameters in studying reservoir properties and compartmentalization in both conventional and unconventional energy resources. Fractures could also play a significant role in the assessment of subsurface carbon capture storage. Subsurface fracture is often challenging to delineate and map due to its complex origin and paragenesis. This is compounded by time consuming and biased-prone fracture analysis when using conventional methods. In this study, we propose a new approach by coupling simple image binarization and advanced image-to-image translation with Conditional Generative Adversarial networks (CGAN) to fully automate and optimize fracture characterization and delineation processes. Here, fracture maps were generated from both outcrop and subsurface examples and the results show the proposed method is significantly more superior than conventional methods reaching up to 94% in mean accuracy. We further extends the application of our newly proposed method to solve inverse problems in fracture reservoir by generating realistic fracture map only from a simple sketch. This work further highlights the enormous potential of deep learning-assisted analysis on different geological problems, such as fracture characterization and delineation. Understanding reservoir fracture distribution and properties in both 2D and 3D spaces would help to optimize exploration, production, and reservoir monitoring of energy resources.
Original languageEnglish (US)
Title of host publicationSecond International Meeting for Applied Geoscience & Energy
PublisherSociety of Exploration Geophysicists and American Association of Petroleum Geologists
Number of pages4
StatePublished - Aug 15 2022
Externally publishedYes

Bibliographical note

KAUST Repository Item: Exported on 2023-03-02
Acknowledgements: We thank Dr. Hussein Hosseit from the King Abdullah University of Science and Technology for the dataset provided.
This publication acknowledges KAUST support, but has no KAUST affiliated authors.


Dive into the research topics of 'A novel deep learning-assisted reservoir fracture delineation with Conditional Generative Adversarial Networks'. Together they form a unique fingerprint.

Cite this