Q2. What are the future works mentioned in the paper "Multi-layer domain adaptation method for rolling bearing fault diagnosis" ?
The two issues mentioned above will be focused on in further research, as well as the optimization of the hyper-parameters in the proposed method.
Q3. Why is the principal component analysis first adopted?
The principal component analysis (PCA) is first adopted to reduce the dimensionality of the feature data to 50 and suppress signal noise.
Q4. How long is the average computing time for the network training?
Since the network training process is implemented off-line, the longest average computing time of 865.5 seconds for 2000 epochs in this case is still acceptable in the proposed fault diagnosis framework.
Q5. What is the way to improve the classification accuracy of the proposed method?
If the enhanced experimental settings of the proposed method is used regardless of the off-line computational burden for training, higher classification accuracy can also be achieved.
Q6. What are the main components of rolling element bearings?
Rolling element bearings are critical components in heavy-duty machineries, manufacturing systems etc. and have been widely applied in modern industries.
Q7. How high is the cross-domain testing accuracy?
As high as 99.17% cross-domain testing accuracy is obtained with the default experimental setting, and up to 99.76% accuracy can be achieved using the enhanced network configuration.
Q8. What is the proposed deep learning method?
In general, the proposed deep learning method combines two architectural ideas for better feature extraction of vibration signals, i.e. CNN and fullyconnected layer.
Q9. What are the vibration signals used in this study?
The vibration signals used in this study were collected from the drive end of the motor in the test rig on four different health conditions: 1) normal condition (H); 2) outer race fault (OF); 3) inner race fault (IF); and 4) ball fault (BF).
Q10. How many bearing health conditions were considered in the case study?
In the latter case, 95% and higher testing accuracies were achieved in [62–64] where 4 bearing health conditions or fewer were considered.
Q11. What is the impact of the introduced punishment factor on the diagnosis accuracy?
The introduced punishment factor α, which determines the domain adaptation strength, may also have influence on the diagnosis accuracy.
Q12. How is the domain shift problem solved?
As shown in Figure 1, the domain shift problem is expected to be solved by jointly minimizing the classification error and the distribution discrepancy between the source and target domains.
Q13. How many filters are adopted in each layer?
As FN becomes larger, the average testing accuracy increases stably, and it reaches 99.75% when 50 filters are adopted in each layer.