Wintershall Dea primarily conducts two types of AI projects: traditional, large-scale projects and small, easy-to-implement “fireflies.” A firefly is a Wintershall Dea concept for conducting a quick, scalable AI project to solve a simple problem. Since there are employees trained in data science throughout the company, business units can develop and code fireflies independently and call upon the CoC for support as needed.
Fireflies start small, then sometimes catch fire. When they do, they are built to scale quickly. For example, an employee in an engineering department was tasked with manually extracting key values from more than 2,000 PDF documents and feeding that data into a spreadsheet. The process was tedious and took time away from the employee for more creative, meaningful work.
Applying AI, the engineering team was able to automate the process, enabling the employee to work on more challenging projects and providing greater overall value to the company. It soon became apparent that the same model for extracting real-time data from internal and external sources could be valuable to other parts of the business and beyond. Today, the scalable solution is applied in several business and corporate units for a variety of purposes.
Large-scale projects aim high from the beginning. In 2021, Wintershall Dea investigated the application of AI to maintaining the integrity of its gas and oil wells in Norway. That maintenance is particularly important for wells in operation, especially subsea wells. With miles of massive pipes encased in multiple layers of steel and concrete burrowing deep into the seabed, small leaks could be imperceptible for long periods of time even in the presence of state-of-the-art well monitoring system—until they become large enough to cause major problems in the worst cases. Thus, early detection is essential.
Previously, Wintershall Dea engineers had been monitoring data from well sensors on an ongoing basis. But even with day-to-day analysis, some issues were simply indetectable to human beings.
Using AI, the team developed a use case for analyzing data from existing sensors much more intensively and accurately than was previously possible. “We first sought to validate the hypothesis that we could use AI to detect a historical leakage incident,” says Prihandono Aditama, Product Manager at Wintershall Dea. “Once we were able to confirm we had the right model, we connected it with live data from the well sensors.
“Currently, if the AI detects an anomaly, it sends an email to our engineers,” he continues. “We’re in the process of building a user interface for the engineers, which will be available in the first release of the product.”
The IBM AI@Scale tools and methodology have been instrumental throughout the process. “One of the major contributions IBM brings to us is how to take the proof of concept live into production,” says Aditama. “The templates IBM provides enable us to do quick scaling, and to do testing, proofs of concept and development in parallel.”