L2M – DeepFault AI: Advanced Geoscience Software Solution

This project will validate the commercial potential of DeepFault AI, an advanced geoscience software solution that automates seismic fault interpretation using deep learning, with integrated uncertainty quantification and quality control (QC). In energy sectors such as oil and gas, geothermal, and carbon storage, reliable subsurface interpretation is essential. Traditional fault detection methods are slow, subjective, and lack mechanisms to quantify confidence.

DeepFault AI addresses these gaps by combining high-performance fault segmentation with real-time uncertainty estimation, allowing users to assess the reliability of each prediction. This is critical in high-stakes environments where interpretation errors can lead to costly drilling decisions. The system also includes a QC workflow that flags low-confidence areas, helping interpreters prioritize review.

The technical foundation is a deep neural network tailored for seismic data, using Monte Carlo Dropout, concrete dropout, and ensemble methods for uncertainty, along with patch-based training and transfer learning for broader applicability.

Through Lab2Market Validate, we aim to:

Engage industry stakeholders (e.g., oil & gas operators, CCS developers) to validate demand for uncertainty-aware interpretation;

Interview 75+ potential users to assess business viability, value proposition, and pricing;

Evaluate integration with industry tools (Petrel, OpendTect, RMS) for seamless adoption;

Validate use cases where uncertainty/QC maps guide decisions in exploration, reservoir modeling, and fault connectivity;

Assess IP, regulatory, and data management needs for on-prem or cloud deployment.

Deliverables include:

A validated business model and commercialization roadmap

Feedback from end users and early adopters

Prioritized customer segments and use cases

Demonstrated value of uncertainty/QC outputs in practice

This project supports the transition of research into scalable innovation, enabling safer, more automated, and interpretable subsurface workflows.