Based on comprehensive field data covering more than 22,000 operating years of onshore and offshore wind turbines on five different continents, we present failure rates for the main components of the converter system, investigate their failure behavior with respect to trends and identify both design-related and site-specific factors having a significant effect on converter reliability. All converter components show prominent early failure behavior followed by a transition straight to degradation failures. Several design factors are identified as significant reliability drivers: namely the cooling concept of the converter, its position inside the wind turbine, its rated capacity as well as the IGBT-module manufacturer. In addition, a susceptibility to the environmental influences of mean absolute ambient humidity, mean ambient temperature, installation height and mean capacity factor is identified and quantified.

SCADA operating data are more and more used across the wind energy domain, both as a basis for power output prediction and turbine health status monitoring. Current industry practice to work with this data is by aggregating the signals at coarse resolution of typically 10-min averages, in order to reduce data transmission and storage costs. However, aggregation, i.e., downsampling, induces an inevitable loss of information and is one of the main causes of skepticism towards the use of SCADA operating data to model complex systems such as wind turbines. This research aims to quantify the amount of information that is lost due to this downsampling of SCADA operating data and characterize it with respect to the external factors that might influence it. The issue of information loss is framed by three key questions addressing effects on the local and global scale as well as the influence of external conditions. Moreover, recommendations both for wind farm operators and researchers are provided with the aim to improve the information content. We present a methodology to determine the ideal signal resolution that minimized storage footprint, while guaranteeing high quality of the signal. Data related to the wind, electrical signals, and temperatures of the gearbox resulted as the critical signals that are largely affected by an information loss upon aggregation and turned out to be best recorded and stored at high resolutions. All analyses were carried out using more than one year of 1 Hz SCADA data of onshore wind farm counting 12 turbines located in the UK.

The installed wind energy capacity increases every year. However, operation and maintenance costs still make up a considerable portion of the levelized costs of electricity. This costs can be greatly reduced by the application of suitable early fault detection methods. The supervisory control and data acquisition system of wind turbines is one key element, as nowadays it provides high resolution operating data. Such data can be used for the implementation of normal behavior models, one precursor for concise early fault detection methods. In this work, we implement a generally applicable and interpretable partly physics‐based, partly empirical normal behavior model for drive train temperatures in wind turbines. The model achieves a coefficient of determination well above 0.9 for 12 of 16 available temperature signals. We also present the possibility of interpreting the model's underlying physical correlations for a selected temperature signal. In addition, we evaluate the influence of the input data's temporal resolution on the model quality. While models based on 10 s data perform slightly better, the results show that 10 min data are sufficient for modeling temperatures.

Power converters are among the most frequently failing subsystems of onshore and offshore wind turbines. In order to minimize the resulting downtime and production losses, the time to repair should be as low as possible. In practice, however, it is not uncommon for several turbine visits to be necessary, as information about the failure mode and the spare parts required can often only be determined on site. This paper presents a data-driven, interpretable workflow for the remote diagnosis of power-converter–related turbine shutdowns using converter control system data from an offshore wind farm. The study uses converter-fault events and three data sources: high-resolution fast logs (4.5 kHz, −350 ms to +200 ms around a fault-induced trigger), 1-min operating data, and fault flags derived from event log data. From an initial 864 engineered features we remove low-variance and highly correlated features, apply a subsampled decision-tree inclusion-rate filter to retain 34 features, and estimate diagnostic impact via subsampled logistic regression. Results show that fast-log features and converter fault flags contain the most predictive information for classifying standstill severity after a fault-induced shutdown, while low-resolution operating data contribute little. Using four of the derived features yields the best cross-validated performance in a decision tree with an accuracy of 0.89 and an F1-score of 0.86. The proposed approach is practical for industry use and offers the potential to provide explainable decision support for improving first-time fix rate.

Power converters are among the most frequently failing subsystems of onshore and offshore wind turbines. In order to minimize the resulting downtime and production losses, the time to repair should be as low as possible. In practice, however, it is not uncommon for several turbine visits to be necessary, as information about the failure mode and the spare parts required can often only be determined on site. This paper presents a data-driven, interpretable workflow for the remote diagnosis of power-converter-related turbine shutdowns using converter control system data from an offshore wind farm. The study uses converter fault events and three data sources: high-resolution fast logs (4.5 kHz, −350 to +200 ms around a fault-induced trigger), 1 min operating data, and fault flags derived from event log data. From an initial 864 engineered features we remove low-variance and highly correlated features, apply a subsampled decision-tree inclusion-rate filter to retain 34 features, and estimate diagnostic impact via subsampled logistic regression. Results show that fast-log features and converter fault flags contain the most predictive information for classifying standstill severity after a fault-induced shutdown, while low-resolution operating data contribute little. Using four of the derived features yields the best cross-validated performance in a decision tree, with an accuracy of 0.89 and an F1 score of 0.86. The proposed approach is practical for industry use and offers the potential to provide explainable decision support for improving the first-time fix rate.