Simultaneous fault diagnosis of wind turbine using multichannel convolutional neural networks. (February 2021)

• Record Type: Journal Article
• Authors: Zare, Samira; Ayati, Moosa
• Is Part Of: ISA transactions. Volume 108(2021)
• Page Start: 230
• Page End: 239
• Publication Date: 2021-02
• Subjects: WT wind turbine -- CMFD condition monitoring and fault detection -- ANN artificial neural network -- CNN convolutional neural network -- MC-CNN multichannel convolutional neural network -- FAST Fatigue, Aerodynamics, Structures, and Turbulence -- ReLU Rectified linear unit -- GPU graphical processing unit -- NREL National Renewable Energy Laboratory -- SCADA supervisory control and data acquisition -- SNR signal-to-noise ratio; Wind turbine fault diagnosis -- Multichannel convolutional neural networks -- Imaging time-series -- Signal to image conversion; Engineering instruments -- Periodicals; Engineering instruments; Periodicals; Electronic journals; 629.805
• DOI: 10.1016/j.isatra.2020.08.021
• Languages: English
• ISSNs: 0019-0578

Abstract

Abstract: Wind turbine technology is pursuing the maturation using advanced multi-megawatt machinery equipped by powerful monitoring systems. In this work, a multichannel convolutional neural network is employed to develop an autonomous databased fault diagnosis algorithm. This algorithm has been evaluated in a 5MW wind turbine benchmark model. Several faults for various wind speeds are simulated in the benchmark model, and output data are recorded. A multichannel convolutional neural network with multiple parallel local heads is utilized in order to consider changes in every measured variable separately to identify subsystem faults. Time-domain signals obtained from the wind turbine are portrayed as images and fed independently to the proposed network. Results show that the multivariable fault diagnosis scheme diagnoses the most common wind turbine faults and achieves high accuracy. Highlights: Seven fault scenarios for wind turbine subsystems are simulated in FAST. Time-domain signals are fed into the convolutional neural network as digital images. Multichannel convolutional neural network with seven input heads classifies images.