A Novel Method for Remaining Useful Life Prediction of Roller Bearings Involving the Discrepancy and Similarity of Degradation Trajectories

Table 2

Specific setting parameters of the convolutional encoder.

Layer	Parameters	Output size	Activation function
Input layer	—	128 × 128 × 3	—
Convolutional layer 1	15 kernels, size: 3 × 3	128 × 128 × 15	ReLU
Pooling layer 1	Size: 4 × 4, stride: 4 × 4	32 × 32 × 15	—
Convolutional layer 2	15 kernels, size: 3 × 3	32 × 32 × 15	ReLU
Pooling layer 2	Size: 4 × 4, stride: 4 × 4	8 × 8 × 15	—
Convolutional layer 3	15 kernels, size: 3 × 3	8 × 8 × 15	ReLU
Pooling layer 3	Size: 2 × 2, stride: 2 × 2	4 × 4 × 15	—
Fully connected layer	240 neurons	1 × 240	—