Search

The existing supervisory control and data acquisition (SCADA) system continuously collects data from wind turbines (WTs), which provides a basis for condition monitoring (CM) of WTs. However, due to the complexity and high dimension and nonlinearity of data, effective modeling of spatial-temporal correlations among the data still becomes a primary challenge. In this paper, we propose a novel CM approach based on the m ultidirectional s patial- t emporal f eature a ggregation n etworks (MSTFAN) to accurately evaluate the performance and hence diagnose the faults of the turbines. Firstly, the data collected from various sensors are formulated into graph-structured data at each timestamp. Spatial-temporal network constructed by combing a graph attention network (GAT) and a temporal convolutional network (TCN) is used to extract spatial-temporal features of the data. Then, a bi-directional long short-term memory (BiLSTM) neural network is adopted to further study long-term spatial-temporal dependency of the extracted features. Finally, the condition score is obtained and the streaming peaks over threshold (SPOT) is applied to determine the abnormal threshold for early warning of the fault occurrence. Experiments on datasets from real-world wind farms demonstrate that the proposed approach can detect the early abnormal situation of the WTs, and outperform other established methods. [ABSTRACT FROM AUTHOR]

With the applications of Machine Learning (ML) in condition monitoring (CM) of wind turbines (WTs), regression-based approaches are mainly for WT power curve modelling to evaluate its performance. A fitted performance curve (PC) of power versus wind speed is prevailing and straightforward for WT performance monitoring. As WT operations could be affected by multiple parameters. It is difficult to identify the fault type merely by a power curve. Particularly at the rotor side of doubly fed induction generator (DFIG), there are miscellaneous electrical disturbances that have their distinct effects on current signals. Hence, by merely processing the rotor currents, a PC of standard deviation of currents versus wind speed can be employed for CM of rotor-side faults. A non-parametric regression model, Gaussian process regression (GPR), is used to predict the PC under normal operation, and afterwards, Fuzzy Inference System (FIS) is applied for fault diagnosis. Therefore, the paper proposes a hybrid GPR and FIS approach towards condition monitoring (CM) and fault diagnosis at the rotor side of a DFIG by merely processing rotor currents through various signal processing techniques, where miscellaneous electrical disturbances can be recognized and localized. The comparative results show that the hybrid approach has superiority over two common ML classifiers, support vector machine (SVM) and decision tree (DT), for fault diagnosis under both test cases including drift faults and excluding drift faults. By disregarding drift faults, this hybrid approach is validated to be efficient for a real-time application with overall promising performance metrics in terms of accuracy (98.29%), dependability (98.12%), and security (100%). The diagnosed results on individual faults and research shortcomings are also discussed. [ABSTRACT FROM AUTHOR]

This paper proposes a data driven model-based condition monitoring scheme that is applied to wind turbines. The scheme is based upon a non-linear data-based modelling approach in which the model parameters vary as functions of the system variables. The model structure and parameters are identified directly from the input and output data of the process. The proposed method is demonstrated with data obtained from a simulation of a grid-connected wind turbine where it is used to detect grid and power electronic faults. The method is evaluated further with SCADA data obtained from an operational wind farm where it is employed to identify gearbox and generator faults. In contrast to artificial intelligence methods, such as artificial neural network-based models, the method employed in this paper provides a parametrically efficient representation of non-linear processes. Consequently, it is relatively straightforward to implement the proposed model-based method on-line using a field-programmable gate array. [ABSTRACT FROM AUTHOR]