Compressor anomaly detection: two weeks' warning of a dry gas seal failure
An alert two weeks ahead of a dry gas seal failure on an export compressor; 300–370 kboe of production loss avoided on the one failure identified, USD 12–16 million a year in value.
Challenge
Compressor failures are rare events with severe consequences: a single failure can cost tens of millions of dollars in lost production. The control system did not capture the compressor's state well enough to prevent them.
Solution
Arundo built a cloud-based machine learning system for early warning of abnormal compressor behaviour. An unsupervised model learned the behaviour of the export compressors over several years of data and flagged deviations. It raised an alert two weeks before a dry gas seal failure and named the most anomalous sensors.
Impact
300–370 kboe of production loss avoided on the single identified failure; USD 12–16 million a year in value.