Challenge
Monitoring dry seal gas systems on large compressors is complex. Early signs of seal degradation often appear as subtle changes in pressure or flow, which can be difficult to distinguish from normal process fluctuations caused by changing loads and operating conditions.
Traditional monitoring approaches risk either detecting issues too late or generating false alarms that reduce trust in alerts. The challenge was to identify meaningful deviations early enough to prevent seal degradation, while avoiding unnecessary interventions triggered by normal variability in the process.
Solution
Process engineers used historical compressor data to build and validate machine learning models based on known dry seal gas failure events. The models focused on key signals, including primary vent discharge pressure for both drive-end (DE) and non-drive-end (NDE) components.
Once trained, the models were used to monitor new data as it became available and flag patterns that deviated from normal operating behavior. Model outputs are reviewed on a weekly basis together with the integrity team, covering all relevant compressor units.
This setup allowed engineers to combine model insights with their domain expertise to assess developing issues, prioritize follow-up actions, and build confidence in the approach over time.
Value Delivered
- Eliminated irrelevant hypotheses related to the production process
- Significant cost savings in time and resources
- Avoided production downtime for equipment testing
- Resolved the issue, normalizing chlorine gas levels
