Ensemble Dilated Convolutional Neural Network and Its Application in Rotating Machinery Fault Diagnosis

Table 1

Parameters of the dilated convolutional neural network model.

Object			Hyperparameter settings
Feature extractor	Block 1	Convolutional layer #1	Number of channels: 4, kernel width: 5, stride: 2
		Pooling layer #1	Kernel width: 2, stride: 2
	Block 2	Convolutional layer #2	Number of channels: 8, kernel width: 5, stride: 2
		Pooling layer #2	Kernel width: 2, stride: 2
	Block 3	Convolutional layer #3	Number of channels: 16, kernel width: 5, stride: 2
		Pooling layer #3	Kernel width: 2, stride: 2
Decision maker	Block 4	Fully connected layer #1	Network width: input dimension
		Fully connected layer #2	Network width: 128
		Fully connected layer #3	Network width: number of fault category
Early stopping			5
Maximum number of iterations			100
Learning rate			10^−4
Small batch size			100