Technology and experience of identifying tectonic faults based on seismic data using artificial intelligence

UDK: 550.834.05

DOI: 10.24887/0028-2448-2025-11-6-12

Key words: deep learning, neural networks, seismic exploration, tectonic faults, fault detection, image segmentation

Authors: V.D. Grishko (RN-KrasnoyarskNIPIneft LLC, RF, Krasnoyarsk; Siberian Federal University, RF, Krasnoyarsk); A.A. Koziaev (RN-KrasnoyarskNIPIneft LLC, RF, Krasnoyarsk); R.S. Melnikov (Rosneft Oil Company, RF, Moscow); D.A. Petrov (RN-KrasnoyarskNIPIneft LLC, RF, Krasnoyarsk); T.E. Khokhryakova (RN-KrasnoyarskNIPIneft LLC, RF, Krasnoyarsk); A.S. Vostretsov (RN-KrasnoyarskNIPIneft LLC, RF, Krasnoyarsk; Siberian Federal University, RF, Krasnoyarsk)

The article presents a method for the automatic detection of tectonic faults in 3D seismic data using a deep convolutional neural network based on the UNet architecture. A key challenge in applying artificial intelligence to seismic interpretation is the severe scarcity of high-quality labeled training data, as fault labeling in real seismic volumes is subjective, labor-intensive, and often incomplete. To address this issue, the authors propose the use of synthetically generated seismic models, in which fault geometry and parameters are precisely and unambiguously defined during the modeling stage. This approach enables the creation of a large-scale, representative training dataset encompassing a wide variety of fault types and geological settings. To adapt the model to real field data, a fine-tuning mechanism is implemented using a limited set of expert-interpreted seismic sections. The modified multi-level network architecture that ensures high sensitivity to thin and elongated tectonic features and produces a probabilistic fault cube that reflects the model’s confidence in fault presence at each location. Practical testing on real data from Rosneft Oil Company confirmed the high effectiveness of the proposed approach: after fine-tuning, the model demonstrates significantly improved fault detection performance, thereby enhancing the efficiency, objectivity, and reproducibility of the interpretation process. The developed method enables geophysicists to focus on result analysis rather than on the routine task of structural delineation and was proven successful across diverse tectonic settings.

Refereces

1. Wu X., Liang L., Shi Y., Fomel S., FaultSeg3D: Using synthetic data sets to train an end-to-end convolutional neural network for 3D seismic fault segmentation, Geophysics, 2019, V. 84, no. 3, DOI: https://doi.org/10.1190/geo2018-0646.1

2. Fossen H. Structural geology, Cambridge: Cambridge University Press, 2012, 940 p.

3. Wu X., Geng Z., Shi Y. et al., Building realistic structure models to train convolutional neural networks for seismic structural interpretation, Geophysics, 2019, V. 85,

no. 4, DOI: https://doi.org/10.1190/geo2019-0375.1

4. Sheng H., Wu X., Si X. et al., Seismic Foundation Model (SFM): A new generation deep learning model in geophysics, 2023,

DOI: https://doi.org/10.48550/arXiv.2309.02791

5. Zhang C.N., Zhang C.S., Song J. et al., A survey on masked autoencoder for self-supervised learning in vision and beyond, 2022,

DOI: https://doi.org/10.48550/arXiv.2208.00173