A colleague pointed me to an article, Timeline of a refinery pump failure and how it could have been prevented, on the Belgium-based EngineeringNet.be website. The story was about a South American refinery that had a high-speed centrifugal pump fail catastrophically resulting in production losses and large repair costs. Todd Reeves is in Emerson’s Machinery Health Management team, part of the Asset Optimization organization.
What happened was an inboard bearing lost lubrication, overheated and finally seized up. The unfortunate part of the story is an automated motor-pump train monitor and advanced vibration analysis system had been installed four months earlier and was working properly.
This monitoring equipment included the CSI 9210 Machinery Health Transmitter connected to the automation system via Foundation fieldbus. This equipment did its job communicating advisory alarms it began to detect problems in the lubrication system.
These alerts went unheeded until they became maintenance alerts and ultimately failure alerts. Todd wrote that the health curve of the pump deteriorated rapidly in the final ten minutes before failure.
Why? The equipment did its job and dutifully reported the problem. The issue turned out to be more of overall unit tuning and alarm management issues. These alerts had been lost among other alarms coming in.
Working as a team, the refinery and local asset optimization experts reviewed the overall alarm strategy and identified opportunities to reduce the alarms and alerts coming in to the operators.
Specifically for the pumps, a best practice was established to add additional temperature measurements on the pump. Training was established to clarify how these alerts would be transitioned between the operators and maintenance staff. Clarifying this process is important when working with predictive diagnostics. At the time, it is not yet an actual problem–but like this centrifugal pump example–will fail if not addressed.