A New Generative Neural Network for Bearing Fault Diagnosis with Imbalanced Data

Table 4

Configuration of the GCNN.

Layer name	Layer configuration
Convolution layer 1	3 × 3, 32
Max pooling layer 1	2 × 2, stride 2
Convolution layer 2	3 × 3, 64
Max pooling layer 2	2 × 2, stride 2
Convolution layer 3	3 × 3, 128
Max pooling layer 3	2 × 2, stride 2
Convolution layer 4	3 × 3, 256
Upsample 1	3 × 3, 128
Merge 1	—
Convolution layer 5	3 ×v3, 64
Upsample 2	3 × 3, 64
Merge 2	—
Convolution layer 6	3 × 3, 32
Upsample 3	3 × 3, 32
Merge 3	—
Convolution layer 7	3 × 3, 32
Flatten layer	—
Softmax layer	10 (or 4 or 16)