APPLICATION OF A DYNAMIC PROGNOSTIC MAINTENANCE POLICY TO offshore wind turbine farms

Many industries identify condition monitoring as a major opportunity to reduce maintenance costs and increase equipment availability. This is also the case for the fast emerging offshore wind turbine technology. However, when considering offshore wind turbine farms, implementation of condition monitoring also introduces a significant capital investment cost. Moreover, condition monitoring systems (CMS) are not infallible: they could potentially miss out on important failures or generate costly false alarms. This makes that the added value of condition monitoring should be assessed rigorously before implementation. Several publications on optimization of condition-based maintenance for wind turbines already appeared in literature (Besnard and Bertling, 2010; Garcia et al., 2006; Nielsen and Sørensen, 2011; Nilsson and Bertling, 2007). The objective of this paper is to quantify the added value of a prognostic maintenance policy for an offshore wind turbine farm. The prognostic maintenance policy makes use of predictive information, more specifically the remaining useful life of different components of the wind turbine. The remaining useful life estimation is based on measurements of several relevant features (e.g. vibrations), based on the evolution of these features through time a prediction of the remaining useful life of the considered component can be made. Compared to traditional condition-based maintenance policies more information, a prediction of remaining useful life into the future, is available to schedule and optimize maintenance actions. The use of a prognostic maintenance policy is an entire new approach to maintenance scheduling and optimization for offshore wind turbines. An entire wind farm is studied rather than only one wind turbine, which is another major contribution of this paper, in order to introduce the importance of dependencies between the separate wind turbines in the farm. Furthermore, the ability of the prognostic maintenance policy to react to these different dependencies (i.e. economic, structural and stochastic dependence) is quantified.