Remaining Useful Life: Forecasting Wind Turbine Component Failures

Vibration condition monitoring systems (CMS) are widely used in the wind sector to detect the onset of faults, for critical components such as bearings and gearboxes. From the perspective of a wind farm operator, understanding the remaining useful life (RUL) of these components is of great value, helping to maximise component utilisation whilst avoiding costly failures. However, building models that can provide accurate predictions is not easy, since failure signatures are unique to individual failure modes, turbine platforms, component design, operating environments and many other factors. Furthermore, data is often not shared between operators due to commercial sensitivity. These factors often limit the available data for models to learn from, resulting in models which are only valid for very specific cases or with significant predictive uncertainty.

At Onyx Insight, our team is in a unique position to address this challenge: we have access to monitoring data from over 28,000 turbines, across 42 different countries and for 185 different turbine platforms, while our failure database also contains over 30,000 failure events; this wealth of data enables us to better generalise across turbines, sensor hardware types and operating conditions, reducing model uncertainty, and increasing trustworthiness for human decision-makers.

Focused on predicting RUL probabilistically for main bearings and planet bearings, this presentation gives an overview of the problem, and the scalable modelling solutions that we have developed, by combining our extensive condition monitoring and machine learning expertise. The forecasts and actionable insights our models provide could bring substantial savings to wind turbine operations and maintenance activities.