Geoscience · Seismic · Machine learning
Machine learning for geoscientists, without writing code.
GeoXMate takes geological and geophysical data from import to a validated machine-learning model, then applies that model back to wells, seismic volumes and 3D geological models. Each step is a tool the geoscientist controls, not a script to maintain.
Vision
The geoscientist who understands the rocks should be the one who builds, tests and trusts the machine-learning model.
Mission
One workstation where subsurface data is prepared, explored and cleaned, where models are trained, optimised and judged on wells they have never seen, and where the results return to the wells, seismic volumes and geological models they came from.
No code, full control
Every step of the machine-learning workflow is a tool with its own visual checks. The geoscientist decides on the data, the features, the algorithm and its settings, without writing or maintaining scripts.
Validated where it matters
Models are always tested on whole wells held out of training, by blind wells and leave-one-well-out. A score in GeoXMate is a measure of how the model performs in a new well.
One project for every data type
Wells, logs, tops, seismic, interpretations, maps and geological models share one project and one set of units, so data moves between steps without export and re-import.
The workflow
From raw data to a model you can defend
GeoXMate follows the order of a machine-learning study. Every stage keeps its results in the project, so the next stage starts where the last one finished.
- Collect Import wells, logs, tops, seismic volumes, interpretations, maps and geological models into one project.
- Enrich Add derived logs, interpreted horizons and faults, zone logs and zone cubes to the raw data.
- Inspect & clean Visualise the dataset, find gaps and outliers, and record every correction in a reversible recipe.
- Train & optimise Compare algorithms, select the input features and tune the hyperparameters on the same validation.
- Evaluate Score each model on blind wells and by leave-one-well-out, with the full set of regression and classification metrics.
- Apply Predict logs in wells and properties in seismic volumes and 3D geological models.
Data
Every subsurface data type in one project
The quality of a model starts with the data it learns from. GeoXMate gathers raw and interpreted data in one place, converted to the project's units at import, and adds the interpreted products that turn it into training material.
- Well headers, LAS, DLIS and Excel logs, deviation surveys and formation tops
- SEG-Y seismic cubes, horizon and fault interpretations from other software
- Maps, time-depth relationships and complete geological models with their properties
- Derived logs, zone logs from tops and 3D zone cubes built from maps
- A smart correlation panel with assisted well-top picking
Machine learning
Prepare, train, optimise and evaluate
Each stage of model development has its own tool, and all of them validate by well. The geoscientist sees why a model performs as it does, and in which wells.
- Twenty interactive charts to understand a dataset and find its outliers
- A QC scan and a cleaning recipe that never changes the project's data
- Classification and regression with tree ensembles, boosting, kernel methods and neural networks
- Algorithm comparison, feature selection and Bayesian hyperparameter tuning
- Blind-well and leave-one-well-out evaluation with R², RMSE, confusion matrix and more
3D seismic
Interpret horizons and faults with supervised and unsupervised learning
Machine learning in GeoXMate is not limited to well logs. The seismic volume becomes training data for 1D datasets along wells, or for 3D deep-learning networks working on the cube itself.
- Interpreter-guided horizon picking and tracking with fault barriers
- ML Horizon Tracking, trained on interpreted lines and scored on blind lines
- Seismic Event Classes: unsupervised seismic facies along every reflection
- Unsupervised Fault Prediction from the seismic alone, and Supervised Fault Prediction trained on your sticks
- Seismic attributes and trace-to-log conversion for 1D and 3D datasets
3D geological models
Distribute properties with geostatistics or machine learning
Alongside the conventional modelling workflow, GeoXMate predicts properties in every cell with machine learning, using the seismic and the model's own geometry as inputs.
- Import corner-point models with their properties, or build them from maps and faults
- Upscaling of well logs and resampling of seismic cubes into the cells
- Facies and property distribution by kriging, Gaussian and indicator simulation
- Machine-learning distribution trained on the cells, or with models already trained on wells
- Volumetrics with Monte Carlo ranges
GeoXMate Assistant
Ask your project in plain words
The Assistant turns a question such as "Which well has the most GR coverage?" into the right table and chart, and points to the part of the manual that explains the tool to use next.
- Questions about the wells, logs, tops, datasets, models and seismic in the open project
- Answers with the tables and charts behind them, computed by GeoXMate from the project
- How-to questions answered from the built-in manual
- A local AI model on your computer: no internet, and nothing leaves the machine
Shown on open data
The screenshots on this site were made with GeoXMate on the Volve field dataset, released by Equinor and the former Volve licence partners under the Equinor Open Data Licence.
Bring machine learning to your own data
Download GeoXMate, send the licence request from the app with your computer ID, and you will receive a licence file by email.