An Unsupervised Deep Feature Learning Model Based on Parallel Convolutional Autoencoder for Intelligent Fault Diagnosis of Main Reducer

Table 2

The architecture and parameter values of the proposed model.

Model	Original vibrational signals
	Reshaped images in time domain		Spectrogram in time-frequency domain
Parallel feature learning model based on PCAE	Conv1	16 × 12 × 12	Conv1	16 × 12 × 12
	BN1	—	BN1	—
	Down-samp1	Pool size = 2 × 2	Down-samp1	Pool size = 2 × 2
	Dropout1	0.5	Drop1	0.5
	Conv2	32 × 6 × 6	Conv2	32 × 6 × 6
	BN2	—	BN2	—
	Down-samp2	Pool size = 2 × 2	Down-samp2	Pool size = 2 × 2
	Conv3	64 × 6 × 6	Conv3	64 × 6 × 6
	Dropout2	0.5	Drop2	0.5
Fault classifier based on DNN	Flatten layer		—
	Full connection layer 1		32
	Full connection layer 2		16
	Full connection layer 3		7
	Softmax		N/A
	Output		Predicted vector