A Bridge Structure 3D Representation for Deep Neural Network and Its Application in Frequency Estimation

Table 2

3D-CNN configuration.

Input layer	Girder’s tensor	1^st pier’s tensor	2^nd pier’s tensor	18 design parameters
Stage 1	6 × Conv (1,1,32)	6 × Conv (1,1,32)	6 × Conv (1,1,32)	—
	Conv (1,2,32)	Conv (1,2,32)	Conv (1,2,32)
	Dropout (0.3)	Dropout (0.3)	Dropout (0.3)
Stage 2	6 × Conv (1,1,64)	6 × Conv (1,1,64)	6 × Conv (1,1,64)	—
	Conv (1,2,64)	Conv (1,2,64)	Conv (1,2,64)
	Dropout (0.3)	Dropout (0.3)	Dropout (0.3)
Stage 3	6 × Conv (1,1,128)	6 × Conv (1,1,128)	6 × Conv (1,1,128)	—
	Conv (1,2,128)	Conv (1,2,128)	Conv (1,2,128)
	Dropout (0.3)	Dropout (0.3)	Dropout (0.3)
Stage 4	6 × Conv (1,1,256)	6 × Conv (1,1,256)	6 × Conv (1,1,256)	—
	Conv (1,2,256)	Conv (1,2,256)	Conv (1,2,256)
	Dropout (0.3)	Dropout (0.3)	Dropout (0.3)
Flatten part	Flatten layer	Flatten layer	Flatten layer	—
Concatenate part	Concatenate layer
FC part	Dense (256)	Dense (256)	Dense (256)	Dense (256)	Dense (256)
Output layer	1^st order frequency	2^nd order frequency	3^rd order frequency	4^th order frequency	5^th order frequency

Note. 6 × Conv (1, 1, 32) means 6 3D convolution blocks, each block including a 3D convolution layer, a batch normalization layer, and a ReLU activation layer. The convolution size is (1, 1, 1), stride is (1, 1, 1), and the out channel is 32; Conv (1,2,32) means a convolution layer with a convolution kernel of (1, 1, 1), a step of (2, 2, 2), and an out channel of 32; Dropout (0.3) means a dropout layer which randomly sets input units to 0 with a frequency of 0.5 at each step during training time; Dense (256) means a dense layer of 256 units.