A Fault Diagnosis Method of Wind Turbine Blades Based on Improved CNN
A lightweight-improved VGG-19 model based on wavelet transform,depth-separable convolution and convolutional block attention mechanism module is proposed,aiming at the problem that wind turbine blade images with low image resolution can lead to reduced accuracy and speed in the fault diagnosis process;DB4 wavelet and morphology-based enhancement techniques are used to improve the quality of the wind turbine blade images;then the traditional convolutional layer in VGG-19 is replaced with a depth-separable convolutional layer to reduce the number of network parameters and improve the training efficiency;and finally the Convolutional Block Attention Module(CBAM)is introduced to improve the fault diagnosis of wind turbine blades.The results show that the accuracy of the proposed model is 93.91%,the main traditional Convolutional Neural Networks(CNN)models are LeNet,AlexNet,GoogleNet,ResNet-50 and VGG-19,and the proposed model improves over them by 15.06%,8.57%,3.10%,-1.13%and 7.13%;the test time is 0.046 seconds per image,which is a reduction of-0.004,-0.002,0.006,0.015,and 0.010 seconds per image,respectively;the model is lightweight in structure and has higher accuracy and faster detection speed compared to other traditional CNNs.
NETL
NSTL
万方数据
