A correlation-graph-CNN method for fault diagnosis of wind turbine based on state tracking and data driving model

Abstract

Due to the multidimensional parameters and the weak correlations with faults, it is still a challenge for the fault diagnosis of wind turbine based on the large amount of supervisory control and data acquisition (SCADA) data. In this work, a correlation-graph-convolutional neural network (CNN) method was proposed to develop a new methodology to predict the faults of wind turbine. We proposed a state tracking strategy by setting all the 24 h SCADA data before fault emergence as the input data. Meanwhile, the state parameters of wind turbine under normal status were also considered for comparison. Pearson correlation coefficient was conducted to quantitatively calculate the coupling level between state parameters, and a hotspot graph was designed to aggregate the correlation coefficients of all the state parameters to reconstruct an image that was adopted as the input data of CNN model. Finally, two practical faults were employed as the samples to study the performances of the correlation-graph-CNN method. The results showed that the accuracy of fault diagnosis can reach about 90%, that was greatly valuable in practical applications. Hence, our proposed method would have the important contribution on fault diagnosis of wind turbine.