Woodside’s Russell Potapinski has given a resounding endorsement to IBM’s Watson as a means of leveraging its engineering document and data resource. Woodside’s cognitive capabilities build on IBM’s Jeopardy-winning artificial intelligence technology. Following implementation of the Watson Engagement Advisor for linguistic analytics, Woodside deployed Watson Explorer to expose its document corpus as a body of accessible knowledge for the enterprise. Explorer extracts insights from decades of engineering and geological knowledge.
Speaking at the 2016 World of Watson event in Las Vegas last month Potapinski reported on three out of the twelve Watson instances that Woodside has deployed. First up was Watson for Projects where some 33 thousand engineering and drilling documents have been ingested. These were analyzed using a human-machine reinforced learning process conducted by Woodside’s subject matter experts, some of whom have put back their retirement to train the system.
The system can field high level questions such as ‘what are the lessons learned from the Vincent phase III drilling campaign?’ In training, the system returns multiple answers which are ranked by the SME and used to ‘teach’ Watson.
The Explorer now provides pertinent insights for engineers and is used to onboard new employees. As project teams disband and reform, Watson’s ‘memory’ means that lessons learned really get applied. Incidentally, it would take five years to read all the documents in the projects instance.
Watson for drilling likewise has evaluated all prior art, chiefly the well completion reports. The system now understands the relationship between inflows, kicks, stuck pipe and other drilling upsets. It used to take Woodside’s drillers weeks to go through all relevant reports from its own and partner-operated wells. W4D’s GIS interface lets well planners zoom in on an area of interest to show geohazards immediately and to analyze kick risk probability ‘in seconds.’
Watson does not replace humans. But while humans are ‘lousy’ at reading thousands of reports, machines are lousy (for now at least!) at figuring what to do with the information. Putting the two together has changed Woodside’s way of working.
But how do you interact with dozens of Watson instances that blend information from multiple systems? The answer is Potapinski’s pièce de résistance, ‘Willow,’ a bi-directional natural language interface to Woodside’s Watson instances. Willow not only answers questions but displays pages of relevant documentation and even a graph of ‘the evidence,’ tracing the logic and semantics from its response back to the original information sources. Pretty nifty stuff! Watch the video here.
© Oil IT Journal - all rights reserved.