The current development in sensor technology combined with improvements in systems for collecting, storing and analyzing large amounts of data, often associated with the term Industry 4.0, offers the opportunity to identify a larger proportion of faults before they turn into failures. A more proactive maintenance strategy has the potential to reduce maintenance costs by allowing maintenance organizations to focus resources on the right equipment at the right time, and to improve safety and availability by reducing the level of unplanned corrective maintenance. This paper explores the possibilities for predictive maintenance on a set of centrifugal pumps used at an offshore oil platform. As a basis for the analysis, sensor data and maintenance records for 15 centrifugal pumps collected over a period of four years is used. The data is split into a training and a test dataset. Causal tree diagnostic modelling is used to establish the link between failure mode and symptoms for one selected fault, impeller damage. Remaining useful life predictions (RUL) for impeller damage is developed based on a stochastic approach. The paper ends with a discussion of how the insights from the analysis can be used to improve maintenance performance.