Research on Wind Turbine Unbalance Fault Diagnosis Based on Wavelet Transform and Convolutional Neural Network

Rotor imbalance in wind turbines presents a serious problem. Particularly for offshore wind turbines, aerodynamic imbalance could have a severe impact because of the large size of the rotor. A diagnosis method based on a parallel convolutional neural network with multi-scale feature fusion is propos...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access 2024, Vol.12, p.176259-176269
Main Authors: Li, Zhenling, Gao, Yukun
Format: Article
Language:English
Subjects:
Accuracy
Convolution
Convolutional neural network
Convolutional neural networks
Feature extraction
Kernel
Load modeling
rotor imbalance
Rotors
Vibrations
wavelet transform
Wavelet transforms
wind energy
wind turbine
Wind turbines
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Rotor imbalance in wind turbines presents a serious problem. Particularly for offshore wind turbines, aerodynamic imbalance could have a severe impact because of the large size of the rotor. A diagnosis method based on a parallel convolutional neural network with multi-scale feature fusion is proposed to diagnose rotor imbalance. It consists of two feature extractors of different scales, which are combined in the fully connected layer. Firstly, a model of a 3MW wind turbine is built and the mass imbalance and aerodynamic imbalance are added to the simulation. The signal is collected and the effects of rotor imbalance on the nacelle vibration in wind turbines are investigated and described. Secondly, the nacelle vibration is selected as the target signal. Wavelet transform is performed on the collected signals, and the 2-dimensional time-frequency map is obtained as the object dataset for the classification. Thirdly, a convolutional neural network is used to classify rotor imbalances of different magnitudes, and different convolution kernels and activation functions are tested. Finally, a new data set is built in the highly fidelity simulation model, and the trained model is loaded for test and verification. The experiments show that the proposed diagnosis model based on the time-frequency map of nacelle vibrations and a convolutional neural network can identify rotor imbalance effectively, and the accuracy is greater than 98%. The results demonstrate the satisfactory performance of the proposed method. It can diagnose rotor imbalance effectively without additional sensors.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2024.3496921