Bearing Fault Diagnosis Based on Frequency Subbands Feature Extraction and Multibranch One-Dimension Convolutional Neural Network

Table 6

CNN branch LH.

Layer	Filters	Kernel size	Stride	Output size	Padding
Convolution; ReLU	16	24 × 1	3 × 1	86 × 16	Same
Average pooling; Dropout (0.3)		6 × 1	2 × 1	41 × 16	Valid
Convolution	32	4 × 1	1 × 1	41 × 32	Same
Average pooling		2 × 1	2 × 1	20 × 32
Max pooling		8 × 1	1 × 1	13 × 32	Valid
Convolution; ReLU	64	3 × 1	1 × 1	13 × 64	Same
Max pooling		2 × 1	2 × 1	6 × 64	Valid