Smart Engineering Apprentice—AI for rod pump failure analysis

USC has successfully tested machine learning. SEA to be deployed at Chevron’s mid continent unit.

The University of Southern California Viterbi School of Engineering’s CiSoft Smart Engineering Apprentice (SEA) project is applying artificial intelligence to predict rod pump unit, using maintenance strategies as captures from domain specialists. These experts record their operations and maintenance actions along with the ‘signature’ of the rod pump’s status. Computer learning compares combinations of signatures, actions and outcomes to develop rules that can be used to predict approximate times of failure for rod pumps that show signs of impaired functioning. Prof Raghu Raghavendra ventured, ‘In a sense, the computer system is the apprentice of field experts, learning from their past experiences in rod pump maintenance.’

The data captured acts as a historian of rod pump failures and can be leveraged in templates for future repairs. New employees rely on the system as a problem solving ‘point of reference.’

The technique has been trialed on 391 wells in Chevron’s West Texas McElroy using 18 months of history. The system identified 205 pumps as ‘normal,’ and predicted that 47 were on the point of failure. In reality, 11 of the 205 ‘normal’ pumps failed. And of the 47 ‘failures,’ 4 turned out to be normal for an overall 94% correct prediction rate. A second trial identified a further 52 pumps as about to fail. Actually 13 of these worked OK during the subsequent observation period—still providing an 80% predictive accuracy.

Chevron’s Mid Continent Alaska business unit’s ‘iField’ project is behind the SEA R&D. Chevron’s Lanre Olabinjo said, ‘We recognized the potential of the SEA for failure prediction on rod pumping systems and have been evaluating the technology since 2009.’ The first production version of SEA is scheduled for release in October 2011. SEA will be integrated with the Artificial Lift Systems Optimization Solutions already developed for the MCA unit’s Well Performance Decision Support Center (DSC) in Midland Texas. More from

Click here to comment on this article

Click here to view this article in context on a desktop

© Oil IT Journal - all rights reserved.