Presentations

Abstract

Operation and Maintenance (O&M) costs account for 20–30% of the total Levelized Cost of Energy (LCOE) in wind energy, with failures of major components such as gearboxes, generators, and main bearings contributing disproportionately. Accurately characterizing failure rates is challenging due to the low frequency of such events, which necessitates aggregating data across many sites to reach statistical significance. However, this inevitably introduces strong heterogeneity in wind and climate conditions. Since local wind regimes directly govern operational loads and fatigue-driven degradation, neglecting environmental variability represents a critical blind spot in most existing reliability models, limiting their predictive capability and site transferability. This study extends previous work on gearbox reliability by investigating the influence of climate conditions and turbine characteristics on generator and main bearing reliability. The objective is to identify key drivers of component degradation and to develop climate-aware reliability models that can be adapted to site-specific operating conditions. We leverage EPRI’s WinNER collaborative database covering approximately 90 GW of failure data to build one dataset per component, each covering up to 10,000 turbines and approximately 70,000 turbine-years of operation. Observed average failure rates range between 5% and 10%. Environmental covariates are collected from open datasets, including wind speed from the Global Wind Atlas (GWA) and ERA5 reanalysis data, as well as air density, ambient temperature, and relative humidity from ERA5. Siting parameters such as turbulence intensity and shear exponent are obtained from DTU’s Global Atlas of Siting Parameters. Turbine-specific commercial power curves are used to compute capacity factors based on site wind conditions. We employ a Bayesian hierarchical accelerated failure time (AFT) Weibull model. The AFT formulation assumes that covariates accelerate or decelerate component aging, consistent with fatigue-driven degradation mechanisms. The hierarchical structure captures manufacturer-to-manufacturer variability, while a frailty term accounts for unobserved heterogeneity. Covariate selection is optimized individually for each component by maximizing improvements in log-likelihood while minimizing residual frailty. This framework enables a robust separation between environmental influences and design-related variability. Model performance is rigorously assessed using leave-one-out (LOO) cross-validation, showing significant gains in out-of-sample predictive accuracy and a marked reduction of unexplained variability. Results demonstrate that incorporating turbine-specific and climate covariates significantly improves model performance, as evidenced by substantial gains in log-likelihood and reduced unexplained variability. Key drivers of component lifetime include capacity factor, rotor diameter, turbulence intensity, and shear exponent, highlighting the strong coupling between operational loading and environmental exposure. Marked variability between manufacturers is observed, validating the hierarchical approach. Importantly, the framework enables the isolation of climate effects from design-related differences, allowing for consistent estimation of covariate impacts across suppliers. These findings confirm the relevance of climate-aware modeling for predicting component aging under diverse operating conditions. By extending climate-aware reliability modeling from gearboxes to generators and main bearings, this work demonstrates a significant improvement in failure rate characterization. The results enable site-specific reliability forecasting, supporting more accurate long-term O&M planning, optimized maintenance strategies, and improved asset management decisions.