Knowledge Capture and Transfer: Preserving Technical Experience in Geoscience Deep Learning Workflows

  • Blog
  • >
  • Knowledge Capture and Transfer: Preserving Technical Experience in Geoscience Deep Learning Workflows

Scotty Salamoff, Global Geoscience Product Manager and Geophysical Advisor at Bluware

The demand for accurate and efficient seismic interpretation continues to be essential for operators to find ways to reduce risk, increase interpretation effectiveness and throughput, and optimize field production more effectively. However, the oil and gas industry has continued to face a concerning trend – a decline in geoscience specialists equipped with years of invaluable expertise. As skilled and experienced interpreters retire and the number of new entry-level interpreters decline, knowledge that is only obtained through experience is in danger of being lost, a risk that most interpretation software tools cannot help alleviating.

With this massive shift in experienced knowledge base, how will E&P operators tackle this challenge of capturing and preserving geoscientist’s expertise? And more specifically how will interpretation knowledge about certain regions be captured and preserved? How will a newly hired geologist or geophysicist learn how to interpret data at the same level of detail as an expert with 30+ years of experience?

Companies are investigating a wide variety of solutions to help bridge this gap, but traditional interpretation tools primarily used in the industry, such as SLB’s Petrel, IHS Kingdom, and Halliburton’s DecisionSpace, fall short of capturing and applying user experience due to limited or no interactive component to their respective integrated deep learning processes. In addition to traditional tools, this concern also exists in model-centric AI seismic interpretation tools that do not keep the human at the center of the training process.

Bluware’s InteractivAI seismic interpretation tool represents a paradigm shift in the way AI models are trained and deployed. Rather than relying solely on algorithmic optimizations and model-driven approaches (which are commonly deployed in other deep learning tools), this platform empowers interpreters to actively engage with the training process, thereby infusing their valuable human insights and domain expertise into the network learning model.

The real benefit of the interactivity in InteractivAI in deep learning is the direct, real-time collaboration the interpreter has with the AI model. Interpreters provide their expertise through the accuracy and validity of their labels on a small subset of the data to define the training set (usually less than 1%). The model then generates 3D predictions on the entire data set twice per training iteration, providing feedback as visual = predictions. The interpreter confirms or denies the predictions through label association and refinement until the network consistently produces geologically valid results. This feedback loop fosters collaboration between humans and machines, allowing the AI model to learn from the interpreter’s experience, and empowering interpreters with the ability to expand and refine their predictions accordingly.

Bluware InteractivAI effectively preserves and leverages interpreter’s knowledge in several ways:
Contextual Understanding: Human interpreters possess a deep understanding of geology in specific regions. By interacting with the machine learning model during the training phase, the interpreter can convey this contextual knowledge instead of waiting until after the training is finished like in other black box AI approaches.

Pattern Recognition: When fault style or reservoir responses are known by the expert interpreter, InteractivAI can identify that same pattern within the entire data set. In return the interpreter does not have to spend an exorbitant amount of time manually interpreting the data.

Error Correction and Fine-Tuning: Interpreters can identify and correct mispredictions made by the network during the training process, again through the mechanism of label optimization. This iterative feedback loop enables the model to learn from its mispredictions, improving its predictive accuracy and reliability.

Knowledge Transfer and Continuity: Interpreters can even save their InteractivAI labels, network, and entire sessions, allowing for seamless transfer of knowledge of specific regions within their organization to less-experienced interpreters.

As geoscience teams become smaller, they are required to become nimbler and rely on tools that enable them to generate thorough seismic interpretations faster and more geologically accurate. Interactive deep learning exemplifies a groundbreaking approach to achieving knowledge transfer using a balanced combination of human and artificial intelligence.

Notify of
Inline Feedbacks
View all comments

Sign up for updates.