Fraunhofer Institute for Industrial Mathematics ITWMIn the »DLseis« (Deep Learning for Seismic Applications) project, we develop and train neural networks for various steps in the seismic data processing and interpretation workflow. Our focus is on applications that significantly simplify daily data processing work through a high degree of automation. This largely eliminates the need for time-consuming parameterization of classical methods, the visualization of intermediate results, or manual steps – such as defining spatially variable mute functions.
Projekt DLseis – Deep Learning for Large Scale Seismic Applications
Deep Learning Tools for Seismic Applications
Deep Learning (DL) is a groundbreaking technology also for the earth sciences. The project Deep Learning for Large Seismic Applications (DLseis) deals with basic research up to ready-to-use tools.
© Fraunhofer ITWM / freepik
Neural Networks for Seismic Data Processing and Interpretation
Project »DLseis«: Deep Learning for Large-Scale Seismic Applications
Our approach is highly application-oriented: We integrate the trained networks into existing software packages as quickly as possible. We primarily rely on supervised learning – a machine learning method in which networks are trained using known input and output data – and use only synthetic data for this purpose. Thanks to the wide availability of real-world datasets, we focus in particular on applications for post-migration data. Here, we benefit from synergies with our migration software GRT as well as the post-processing methods we have developed. This allows us to further develop the networks in a targeted manner from the developers’ perspective, as well as evaluate the quality of their results from the user’s perspective and identify further optimization needs.
For our project partners, we integrate the developed networks into the »SR (PreStack Pro)« software from Sharp Reflections. This allows experts to test and evaluate the networks using datasets they are familiar with even while the research is ongoing.
Current Research Focus Areas
ML Data Preprocessing After Migration
We are currently expanding all our methods to include 3D gathers derived from onshore seismic data and marine OBN acquisitions. Our focus areas include:
- Removal of Multiples (Figure 2)
- Correction of Trimstatics (Figure 1)
- Widening of Bandwidth
- Elimination of Linear Noise
Figure 1: From left to right: Seismic gathers before alignment; Aligned with classical method. Aligned by ML-Align with DNN trained with an incomplete synthetic training data set. Aligned by ML-Align with DNN trained with an improved synthetic data set.
ML Automated Well Tie
- DNN-based wavelet extraction with automatic time-to-depth mapping using bulk shift, stretch-and-squeeze, and log editing (Figures 3 and 4)
- Expansion to multi-well tie workflows
ML Fault Detection
- Searching for geological faults in seismic data
- Modeling realistic synthetic datasets based on simulated migration
If you are interested, we can provide you with the complete research program.
Figure 2: Left: seismic input gathers before demultiple, right: gathers after ML-demultiple.
Figure 3: From left to right: Well log converted to vertical time (acoustic impedance), reflectivity, synthetic trace (5 copies displayed) modelled for extracted wavelet and reflectivity function, trace of real seismic along well path, correlation between synthetic and real seismic.
Figure 4: Wavelet (left column) associated with pairwise input of reflectivity (central column) and seismic trace (right column).
As a communication format between Python-driven network design and training – carried out on our in-house GPU cluster – and the application to real field data, we use ONNX (Open Neural Network Exchange). The tasks of seismic multiple removal and trimstatics correction are thus reduced to feeding the data to be processed into the ONNX network trained for the respective application. The desired results are simply obtained by extracting the data from the last network layer. Such data processing is completely parameter-free!
Previous Project Partners and Acknowledgements
The following energy companies supported the project at various stages. We would like to express our sincere gratitude for their financial support, the data sets they provided, and their participation in constructive and motivating project meetings.
- BP
- ConocoPhillips
- DNO
- Equinor ASA
- Exxon Mobil Corporation
- Hess Corporation
- MOL-Norge
- OMV
- Petrobras
- VårEnergi
- Wintershall Dea
Project and Participation
The »DLseis« project has been running as a consortium project since 2019. New partners can join at any time for a renewable period of one or two years. During their participation, they receive full access to our research results – including the synthetic datasets. The high quality of these datasets contributes significantly to the success of the trained networks and their excellent transferability to real-world data.
This page is currently under revision. More detailed information is available from July 1, 2026.
Until then, we’d be happy to answer your questions personally. Please contact us if you’d like to learn more about the project concept, the research content, or current results.