A major oil company sought to improve effectiveness of its equipment maintenance spend. Compressor failure accounted for a significant share of the company’s lost production, resulting in a strict routine maintenance policy on all gas compressors. Non-productive time dropped significantly, but planned maintenance events rose significantly. Management sought to implement a data-driven, condition-based maintenance approach where preventative maintenance is prioritized based on real-time data streaming from the equipment.
Arundo’s team of data scientists worked with the customer to implement a condition-based monitoring solution for gas compressors. Arundo developed a machine-learning model to map real-time hydrocarbon readings against a standardized set of operating ranges and deployed the model within the Arundo platform. Given the variance of individual signals, a cluster-based approach was used to aggregate multiple individual signals and eliminate white noise. When sensors detected a deviation from the predicted operating range, the system immediately alerted the user to potentially abnormal equipment behavior. Prior to implementation in the production environment, the system was validated against anonymized historical data. The system successfully identified anomalies related to pending failure at the sensor level several months before the existing control system raised any alarms.