Optimized Fault Diagnosis Algorithm under GAN and CNN Hybrid Model

Table 2

Improved generative adversarial network model parameters.

Neural network name	Layer (operation)	Input format	Output format
Generator network	Fully connected layer	[None,128]	[None,256]
	Batch normalization (batch_norm)	[None,256]	[None,256]
	Nonlinear activation (ReLU)	[None,256]	[None,256]
	Deconvolution layer 1 (conv2d_transpose)	[None,2,2,64]	[None,4,4,16]
	Batch normalization (batch_norm)	[None,4,4,16]	[None,4,4,16]
	Nonlinear activation (ReLU)	[None,4,4,16]	[None,4,4,16]
	Deconvolution layer 2 (conv2d_transpose)	[None,4,4,16]	[None,8,8,1]
	Nonlinear activation (tanh)	[None,8,8,1]	[None,8,8,1]
Discriminator network	Convolutional layer 1 (conv2d)	[None,8,8,1]	[None,4,4,32]
	Batch normalization (batch_norm)	[None,4,4,32]	[None,4,4,32]
	Nonlinear activation (LeakyReLU)	[None,4,4,32]	[None,4,4,32]
	Convolutional layer 2 (conv2d)	[None,4,4,32]	[None,2,2,64]
	Batch normalization (batch_norm)	[None,2,2,64]	[None,2,2,64]
	Nonlinear activation (LeakyReLU)	[None,2,2,64]	[None,2,2,64]
	Fully connected layer	[None,256]	[None,7]
	Softmax	[None,7]	[None,7]