Ensemble Augmentation for Deep Neural Networks Using 1-D Time Series Vibration Data
Purpose Deep Neural Networks (DNNs) typically require enormous labeled training samples to achieve optimum performance. Therefore, numerous forms of data augmentation techniques are employed to compensate for the lack of training samples. Methods In this paper, a data augmentation technique named...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer Link
2022
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/35522/1/Ensemble%20Augmentation%20for.pdf http://umpir.ump.edu.my/id/eprint/35522/ https://doi.org/10.1007/s42417-022-00683-w https://doi.org/10.1007/s42417-022-00683-w |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Purpose
Deep Neural Networks (DNNs) typically require enormous labeled training samples to achieve optimum performance. Therefore, numerous forms of data augmentation techniques are employed to compensate for the lack of training samples.
Methods
In this paper, a data augmentation technique named ensemble augmentation is proposed to generate real-like samples. This augmentation method uses the power of white noise added in ensembles to the original samples to generate real-like samples. After averaging the signal with ensembles, a new signal is obtained that contains the characteristics of the original signal. The parameters for the ensemble augmentation are validated using a simulated signal. The proposed method is evaluated by 10 class-bearing vibration data using three Transfer Learning (TL) models, namely, Inception-V3, MobileNet-V2, and ResNet50. The outputs from the proposed method are compared with no augmentation and different augmentation techniques.
Results
The results showed that the classifiers with the ensemble augmentation have higher validation and test accuracy than all the other augmentation techniques. The robustness assessment conducted with noisy test samples and test samples from different loads showed that the classifiers could obtain much higher robustness when trained with samples from ensemble augmentation.
Conclusion
The proposed data augmentation technique can be applied to 1-D time series data to achieve robust classifiers. |
|---|