Resumen: With the growing wind energy sector, the need for advanced operation and maintenance (O&M;) strategies has emerged. So far, mainly corrective or preventive O&M; actions are applied. Predictive modelling, however, is expected to significantly enhance existing O&M; practice. Here, anticipating wind turbine component failures can enable operators to lower the O&M; cost and is particularly useful for wind farms located in remote areas or offshore locations. Previous research has shown that the failure behaviour of wind turbines and their components is highly influenced by the meteorological conditions under which the turbines operate. Hence, there is a significant need for robust models for failure prediction taking into consideration these conditions. Furthermore, solutions need to be found in order to determine the most suitable input variables for enhancing their prediction accuracy. This study uses failure data obtained from 984 wind turbines during 87 operational WT years. Bayesian belief networks (BBN) are trained based on failure records, technology specific covariates, as well as measurements of the environmental and operational conditions at site. Subsequently, the failure events in a wind farm during a period of 36 months are predicted with the BNN, whereas the failure events of six main components are predicted separately. Furthermore, an extensive sensitivity study is carried out to find the model with the highest prediction accuracy for each component. The influence of each meteorological, operational or technical covariate are discussed in detail. The models achieved a very good accuracy and were able to predict the majority of the component failures over the prediction period. Idioma: Inglés DOI: 10.1088/1742-6596/1037/6/062003 Año: 2018 Publicado en: Journal of physics. Conference series 1037 (2018), 062003 [11 pp] ISSN: 1742-6588 Factor impacto SCIMAGO: 0.221 - Physics and Astronomy (miscellaneous) (Q3)