Using Generative AI to Improve Well Integrity Management
Operators with a large inventory of wells completed over several years, even decades, are faced with the challenge of being able to consolidate and visualize them in one place. Being able to do so is an important first step for evaluating well interventions, well integrity risks, and screening wells for multiple use cases such as plug and abandon and carbon capture, utilization, and storage (CCUS) among others. The final state of these wells is captured in a wide range of documents such as end of well reports, well completion reports, etc. Extracting relevant information manually from this large amount of unstructured data is a tedious task that requires significant allocation of resources.
Wellbarrier™ well integrity life cycle solution supports decision making during planning operations, while also providing a structured framework to proactively manage well integrity. The end user of the solution ingests and visualizes wellbore schematics and the associated barrier envelopes from historical wells on one platform. This allows the users to track the status of their barrier elements and risk rank their wells based on their integrity to compare with other wells in their portfolio.
Teams at INNOVATION FACTORI have built a workflow that automates and accelerates this process by leveraging the power of Generative AI, working in tandem with optical character recognition (OCR) engines. A bespoke foundational model is fine-tuned based on data elements that need to be extracted for specific use cases. If the end goal is to create wellbore schematics the model may be designed to extract attributes such as well header information, completion data, formation tops, blowout preventer (BOP) structure, casing data, etc.
The front end visible to the end user is designed to showcase prefilled data entries populated with extracted data, along with references mapping these structured entries to unstructured data from the OCR engine. This intuitive user interface facilitates a rapid quality check before the extracted data is made available for ingestion to the Wellbarrier solution and serves as additional human oversight into the process.
The model has proven to be very useful for extracting relevant attributes from a range of documents, thereby eliminating the need for the engineers having to scour through hundreds of pages of data manually. In a nutshell, the Generative AI model trained by strong domain and data science experts can sift through heavy, unstructured input data to produce a structured output that is accurate, concise, and relevant for the specified use case with a very high degree of accuracy.
Figure 1: Overall process description
The value created by this application of AI to digitize and organize unstructured data is significant. We have the capability to achieve a multifold reduction in the time it takes to populate wellbore schematics in the Wellbarrier solution. Besides the immediate benefit of saving subject matter expert’s (SME) time, such a workflow has a direct impact on accelerating decisions regarding large scale intervention campaigns and the accuracy of these. These decisions could be instrumental in minimizing production downtime, while also aiding proactive well integrity risk management.
For instance, access to the current state of a collection of wells in a structured format enables operators to quickly prioritize them for plug and abandon jobs. Based on local regulations and business constraints, workflows can be built to generate preliminary abandonment plans along with time and cost estimates. Previously, screening wells for plug and abandon and generating high level plans required by experts to manually read completion reports, evaluate risk, and rank wells based on a small number of factors that could be assessed within the available timeframe. With a partially automated screening process, the analysis can be much more comprehensive and complete. Domain experts can focus their efforts on creating more granular plans for selected wells that already have an outline of the design ready for review and validation.
We are seeing a lot of success in use cases where domain expertise is augmented by applying GenAI to unstructured data. As shared in other blog posts, we have built models for analyzing daily drilling reports and well summaries to extract information required to understand major drilling risks such as stuck pipe, downhole losses, tool failures. The ability to quickly adapt to different document formats and large volume is a step change. By coupling these new workflows with our core software solutions, we are making significant impact on improving the speed and accuracy with which we can extract and process contextual data for consumption in our digital simulators.
