Abstract
To investigate problems involving wind turbines that easily occur but are hard to diagnose, this paper presents a wind turbine (WT) fault diagnosis algorithm based on a spectrogram and a convolutional neural network. First, the original data are sampled into a phonetic form. Then, the data are transformed into a spectrogram in the time-frequency domain. Finally, the data are sent into a convolutional neural network (CNN) model with batch regularization for training and testing. Experimental results show that the method is suitable for training a large number of samples and has good scalability. Compared with Back Propagation Neural Network (BPNN), Support Vector Machine (SVM), Extreme Learning Machine (ELM), and other fault diagnosis methods, the average diagnostic correctness rate is higher; so, the method can provide more accurate reference information for wind turbine fault diagnosis.
Highlights
With the growing shortage of energy, wind power generation has become the preferred alternative energy source
This paper presents a fault diagnosis method for wind turbine gearboxes based on a convolutional neural network
This paperspectrum emphasizes a for way achievefeatures, wind turbine diagnosis.neural
Summary
With the growing shortage of energy, wind power generation has become the preferred alternative energy source. In 2014, the study in reference [5] proposed a noise reduction method for the characteristics of strong nonlinear noise in wind turbine vibration signals They constructed effective features based on the information after de-noising and modeled wind turbines using manifold learning algorithms allowing for early weak fault diagnosis. In 2017, aimed at the nonstationary and nonlinear characteristics of wind turbine vibration signals, reference [12] proposed a novel fault diagnosis method based on integral extension load mean decomposition multiscale entropy and a least squares support vector machine. In reference [13], the detection of electrical asymmetry in rotors in wind turbine WT doubly fed induction generators (DFIGs) has been investigated using a test rig under three different driving conditions, and an effective extended Kalman filter (EKF)-based method was proposed to iteratively estimate the fault signature components (FSCs) and track their magnitude.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
