Generator condition monitoring on an unmanned platform
Alerts raised two to four weeks before most generator failures; USD 1–2 million a year less production lost.
Challenge
Failures in the power generation system of an unmanned export platform caused significant production losses every year. Monitoring practices could not detect the failures coming, nor assure redundancy in the system. Failures were rare and the sensor patterns complex, which defeated manual and rule-based monitoring.
Solution
Arundo built a cloud-based machine learning system that monitors the state of the generators and alerts operators when a combination of sensors deviates from normal behaviour. The system raised its alert two to four weeks ahead of the majority of the generator failures it was tested against, giving operators time to act.
Impact
USD 1–2 million a year less production lost to generator failures.