Faults are among the most critical geological features in subsurface reservoirs. From forming structural traps in conventional hydrocarbon plays to introducing drilling hazards in unconventional settings, faults profoundly influence exploration, development, and production decisions. For decades, accurately mapping these features from seismic data has been a time-consuming and error-prone process that relies heavily on a geoscientist’s experience and manual interpretation skills.
Today, artificial intelligence (AI) is transforming this workflow—and subsurfaceAI is at the forefront of that transformation.
The Challenge of Traditional Fault Interpretation
Seismic fault interpretation has traditionally involved painstaking manual digitization across 2D seismic lines and 3D volumes. This process depends heavily on the interpreter’s eye for pattern recognition and is inherently subjective. It’s not only labor-intensive but also difficult to scale across large datasets or to replicate results across teams.
But if you look at fault interpretation for what it is—a pattern recognition task based on visual features in seismic images—it becomes obvious that AI, particularly deep learning, is ideally suited to the job.
How AI Enhances Fault Interpretation
At subsurfaceAI, we’ve developed a robust, AI-driven fault interpretation workflow that dramatically accelerates interpretation while increasing consistency and accuracy. Here’s how it works:
- Synthetic Fault Dataset Generation
We begin by generating a large set of synthetic seismic volumes with embedded fault structures. Users can define the statistical properties of the fault population—such as orientation, dip, length, height, and displacement. These faults are embedded within synthetic seismic volumes that simulate different geological scenarios, including folding and noise. This enables the generation of thousands of training examples efficiently.
- Base Model Training with AI
These synthetic datasets serve as the foundation for training powerful AI fault models. Users can also incorporate proprietary seismic data and previous fault interpretations into the base model training process. This flexibility ensures that the models learn from both synthetic and real-world examples.
- Fine-Tuning for Specific Seismic Surveys
Users can optionally fine-tune base AI models using a few interpreted lines from the target dataset. This step is interactive—users see results instantly as they guide the model to adapt to the specific data characteristics. However, if preferred, a pretrained model from the subsurfaceAI model library can be directly applied without additional tuning.
- AI-Powered Fault Detection
Once trained or fine-tuned, the AI model is applied to the full seismic dataset, generating a fault probability volume (for 3D data) or fault probability sections (for 2D data). These probability maps provide interpreters with an objective and consistent view of likely fault locations.
- Extracting Fault Surfaces
Using the fault probability output, fault surfaces are extracted as 3D mesh objects. These can be exported as fault sticks in standard industry formats and seamlessly integrated into interpretation and modeling platforms.
What Sets subsurfaceAI Apart?
subsurfaceAI’s approach offers several unique advantages over competing solutions:
- Custom Base Model Training: Users can build their own base models tailored to their datasets.
- Leverage Prior Interpretations: Previously interpreted faults can be used as training data.
- Enterprise-Ready AI Management: A sophisticated system for managing labels and models ensures compliance with corporate data governance and facilitates cross-project collaboration.
The Future of Fault Interpretation
By integrating AI into the fault interpretation process, geoscientists can shift from manual, repetitive tasks to higher-value work like validating interpretations, identifying complex structures, and integrating geological understanding. The result is faster turnaround, greater consistency, and ultimately better reservoir understanding.
Interested in learning more or seeing the workflow in action? Visit subsurfaceAI.ca or contact us for a demo ([email protected]).
