Roll-out across plants and sites With the success of this solution in one plant, deployment to all critical compressor discharge flow transmitters was rolled out. For those with a history of failure, a failure model was developed; however, where there have been no instances of choking, an anomaly agent was developed to create alerts if any anomaly is detected. The solution has already been deployed to more than 50 compressor discharge flow transmitters, and it is continuously being deployed to many more compressors. “ By employing an unsupervised machine learning autoencoder algorithm, the model effectively identifies subtle yet critical signs of impulse line choking ” Conclusion The compressor flow transmitter anomaly early event detection model represents a significant advancement in industrial predictive maintenance. By employing an unsupervised machine learning autoencoder algorithm, the model effectively identifies subtle yet critical signs of impulse line choking. Its successful offline and online validations, particularly its ability to detect anomalies well in advance and to correctly interpret complex plant incidents, demonstrate its encouraging accuracy and robustness. This solution not only prevents costly unplanned shutdowns and production losses but also reinforces the plant’s operational safety and reliability, marking a pivotal step towards a truly intelligent manufacturing ecosystem.
were known to be functioning correctly with no impulse line choking. This validation confirmed that the model effectively differentiates between ‘actual sensor malfunctions (impulse line choking)’ and ‘process upsets that affect flow readings but do not originate from sensor malfunction’. It demonstrated that the model is robust and does not produce false alerts during normal operational disturbances not related to impulse line plugging. The online deployment showcased the model’s long-term stability. Trends indicate that the model output consistently remains ‘healthy’, indicating sustained normal operation of the flow transmitters. Benefits and impact Implementation of the ROGC cracked gas compressor third-stage flows anomaly early event detection model offers substantial benefits to the plant: • Advanced alert system : Provides early warnings of anomalous behaviours, specifically impulse line choking, well before they escalate into critical malfunctions. • Proactive maintenance : Empowers stakeholders to undertake timely flushing or cleaning of impulse lines, transitioning from reactive troubleshooting to proactive preventive maintenance. • Avoidance of catastrophic failures : Significantly reduces the risk of unplanned shutdowns and avoids catastrophic failures caused by erroneous flow readings. • Production loss prevention : Directly contributes to saving significant production losses. • Improved efficiency : Minimises downtime and optimises operational continuity. User interface and mail alerts To further enhance user accessibility and integration into operational workflows, a dedicated user interface (UI) was developed. This UI will allow users to easily monitor the status of the flow transmitters and the anomaly detection signal. Also, when any anomaly is detected by the model, an automated email alert is generated to concerned stakeholders, ensuring that relevant personnel are notified upon the detection of any anomaly, facilitating a rapid response.
VIEW REFERENCES
NC Chakrabarti Nc.Chakrabarti@ril.com Priyang Shukla Priyang.shukla@ril.com Jesse Mallhi Jesse.Mallhi@ril.com
Refining India
71
Powered by FlippingBook