Cavitation Analysis in Centrifugal Pumps Based on Vibration Bispectrum and Transfer Learning

Table 1

The structure of the CNN applied in this research.

Stage	Layer no.	Layer type	Size
1	1	Input bispectrum images	227 × 227 × 3
2 (transferred)	2	Convolutional	55 × 55 × 96
	3	ReLU	55 × 55 × 96
	4	Cross-channel normalization	55 × 55 × 96
	5	Max pooling	27 × 27 × 96
3 (transferred)	6	Convolutional	27 × 27 × 256
	7	ReLU	27 × 27 × 256
	8	Cross-channel normalization	27 × 27 × 256
	9	Max pooling	13 × 13 × 256
4 (transferred)	10	Convolutional	13 × 13 × 384
	11	ReLU	13 × 13 × 384
5 (transferred)	12	Convolutional	13 × 13 × 384
	13	ReLU	13 × 13 × 384
6 (transferred)	14	Convolutional	13 × 13 × 256
	15	ReLU	13 × 13 × 256
	16	Max pooling	6 × 6 × 256
7 (trained)	17	Fully connected	10
	18	ReLU	10
8 (trained)	19	Fully connected	5
	20	Softmax	5
	21	Classification output	2