BP - OpMode
Using OpMode to detect failures can result in less unplanned downtime
BP came to Intelligent Plant with data related to one of their choke valves which recently failed. They wanted to see if the issue could have been spotted earlier.
Intelligent Plant utilised our machine learning app, OpMode to try and spot potential choke degradation that occurred on PU-IC.
Choke degradation is an issue that occurs over a long period of time and is not an instance change in process values. Preventing this issue or helping to plan for continuous well operation would save BP millions of pounds through avoiding unplanned downtime.
An example of such savings could be derived from the following results.
Firstly, we had to train OpMode with healthy operating data. Good operational mode was derived from upstream and downstream temperature and pressure tags, as well as the choke position, bounded by a period of several years into its operation.
The change in Principal Components (an interim result tags produced by OpMode) identified a large change 9-10 months before the operations were stopped and the choke was replaced. The application could have alerted about this change and helped engineers to prepare, reducing the likelihood of unplanned downtime.
The above result can be simplified into a binary tag that identifies if the choke is operating in “normal” mode or not.
We can see that over time the choke operation leaves the “good” operational mode more and more often until it was decided to change it in late 2018. Although the binary tag does not identify a single point of failure, we can derive that the application would have indicated that the choke is operated more and more often outside the desirable “good” operational mode. This would have helped engineers to review the operations as well as prepare for potential shutdowns.