Key Learning Outcome: Understand how machine learning and digital twin technology can be used for proactive condition monitoring in subsea systems to enhance performance, reduce interventions, and prevent failures.

All too often, subsea systems experience a controls failure resulting in costly production loss and interventions. At worst case loss of containment can occur. For a long time even the best-in-class operators have mostly used a corrective approach once faults occur and hold a large spares inventory to be able to address current failures. Being able to accurately determine a system/equipment degradation where system components can be replaced before the degradation becomes terminal allows production uptime to be maximised and intervention costs to be minimised by allowing targeted and planned maintenance. Digital twins have long been discussed but have resulted so far in simple ‘traffic light’ go/no-go systems. We present a condition monitoring system that uses machine-learning (ML) derived classification algorithms to report deviations in operational data and identify failing system components. These algorithms are trained using simulations of the system, augmented with faults of varying severities. By simulating faults in isolation, combination, and varying intensities, a large training dataset is generated, enabling the ML algorithms to detect fault “fingerprints” and incorporate them into the classification process. Increasing subsea data availability enables new approaches. An integrated digital twin can remotely monitor system performance in real time and identify trends that may lead to production loss. Autonomous tools can compare live data to the twin, detecting and highlighting deviations from expected performance.