Failures in the main bearings of wind turbines are critical in terms of downtime and replacement cost. Early diagnosis of their faults would lower the levelized cost of wind energy. Thus, this work discusses a gated recurrent unit (GRU) neural network, which detects faults in the main bearing some months ahead (when the event that initiates/develops the failure releases heat) the actual fatal fault materializes. GRUs feature internal gates that govern information flow and are utilized in this study for their capacity to understand whether data in a time series is crucial enough to preserve or forget. It is noteworthy that the proposed methodology only requires healthy supervisory control and data acquisition (SCADA) data. Thus, it can be deployed to old wind parks (nearing the end of their lifespan) where specific high-frequency condition monitoring sensors are not installed and to new wind parks where faulty historical data do not exist yet. The strategy is trained, validated, and finally tested using SCADA data from an in-production wind park composed of nine wind turbines.

As stated by the European Academy of Wind Energy (EAWE), the wind industry has identified main bearing failures as a critical issue in terms of increasing wind turbine reliability and availability. This is owing to major repairs with high replacement costs and long downtime periods associated with main bearing failures. Thus, the main bearing fault prognosis has become an economically relevant topic and is a technical challenge. In this work, a data-based methodology for fault prognosis is presented. The main contributions of this work are as follows: (i) Prognosis is achieved by using only supervisory control and data acquisition (SCADA) data, which is already available in all industrial-sized wind turbines; thus, no extra sensors that are designed for a specific purpose need to be installed. (ii) The proposed method only requires healthy data to be collected; thus, it can be applied to any wind farm even when no faulty data has been recorded. (iii) The proposed algorithm works under different and varying operating and environmental conditions. (iv) The validity and performance of the established methodology is demonstrated on a real underproduction wind farm consisting of 12 wind turbines. The obtained results show that advanced prognostic systems based solely on SCADA data can predict failures several months prior to their occurrence and allow wind turbine operators to plan their operations.

Due to the increasing installation of wind turbines in remote locations, both onshore and offshore, advanced fault detection and classification strategies have become crucial to accomplish the required levels of reliability and availability. In this work, without using specific tailored devices for condition monitoring but only increasing the sampling frequency in the already available (in all commercial wind turbines) sensors of the Supervisory Control and Data Acquisition (SCADA) system, a data-driven multi-fault detection and classification strategy is developed. An advanced wind turbine benchmark is used. The wind turbine we consider is subject to different types of faults on actuators and sensors. The main challenges of the wind turbine fault detection lie in their non-linearity, unknown disturbances, and significant measurement noise at each sensor. First, the SCADA measurements are pre-processed by group scaling and feature transformation (from the original high-dimensional feature space to a new space with reduced dimensionality) based on multiway principal component analysis through sample-wise unfolding. Then, 10-fold cross-validation support vector machines-based classification is applied. In this work, support vector machines were used as a first choice for fault detection as they have proven their robustness for some particular faults, but at the same time have never accomplished the detection and classification of all the proposed faults considered in this work. To this end, the choice of the features as well as the selection of data are of primary importance. Simulation results showed that all studied faults were detected and classified with an overall accuracy of 98.2%. Finally, it is noteworthy that the prediction speed allows this strategy to be deployed for online (real-time) condition monitoring in wind turbines.

This work deals with structural health monitoring for jacket-type foundations of offshore wind turbines. In particular, a vibration-response-only methodology is proposed based on accelerometer data and deep convolutional neural networks. The main contribution of this article is twofold: (i) a signal-to-image conversion of the accelerometer data into gray scale multichannel images with as many channels as the number of sensors in the condition monitoring system, and (ii) a data augmentation strategy to diminish the test set error of the deep convolutional neural network used to classify the images. The performance of the proposed method is analyzed using real measurements from a steel jacket-type offshore wind turbine laboratory experiment undergoing different damage scenarios. The results, with a classification accuracy over 99%, demonstrate that the stated methodology is promising to be utilized for damage detection and identification in jacket-type support structures.

This paper develops a fault diagnosis (FD) and fault-tolerant control (FTC) of pitch actuators in wind turbines. This is accomplished by combining a disturbance compensator with a controller, both of which are formulated in the discrete time domain. The disturbance compensator has a dual purpose: to estimate the actuator fault (which is used by the FD algorithm) and to design the discrete time controller to obtain an FTC. That is, the pitch actuator faults are estimated, and then, the pitch control laws are appropriately modified to achieve an FTC with a comparable behavior to the fault-free case. The performance of the FD and FTC schemes is tested in simulations with the aero-elastic code FAST.