Wasserstein Generative Adversarial Network and Convolutional Neural Network (WG-CNN) for Bearing Fault Diagnosis

Table 5

Comparison of contrast kernel settings.

Experimental group	Number of convolution kernels in different layers			Activation function	Test accuracy
	Layer 1	Layer 2	Layer 3
1	8	16	32	ReLU	0.96
2	16	32	64	ReLU	0.99
3	32	64	128	ReLU	0.98
4	8	16	32	Leaky ReLU	0.95
5	16	32	64	Leaky ReLU	1.00
6	32	64	128	Leaky ReLU	0.98
7	8	16	32	Tanh	0.96
8	16	32	64	Tanh	0.99
9	32	64	128	Tanh	0.97