Imbalanced data fault diagnosis of hydrogen sensors using deep convolutional generative adversarial network with
Yongyi Sun1, Tingting Zhao2, Zhihui Zou2
1Key Laboratory of Electronics Engineering, College of Heilongjiang Province, Heilongjiang University, Harbin 150001, People's Republic of China.
The Review of Scientific Instruments
|October 2, 2021
Summary
This study introduces a novel DCG-CNN method for hydrogen sensor fault diagnosis. It effectively addresses unbalanced datasets by enriching small samples, significantly improving diagnostic accuracy over traditional methods.
Area of Science:
- Sensor technology
- Artificial intelligence
- Machine learning
Background:
- Hydrogen sensor fault diagnosis is critical but challenged by imbalanced data, hindering accuracy.
- Unbalanced datasets in sensor fault diagnosis can lead to unreliable results.
Purpose of the Study:
- To develop an effective method for hydrogen sensor fault diagnosis using deep learning.
- To address the challenge of imbalanced datasets in gas sensor fault diagnosis.
Main Methods:
- A novel deep convolutional generative adversarial network (DCG) combined with a convolutional neural network (CNN) was developed (DCG-CNN).
- 1D sensor signals were converted to 2D gray images to preserve information.
- The DCG model enriched limited fault data samples to create balanced datasets.
- CNN was employed for accurate fault diagnosis, with feature maps visualized to understand performance.
Main Results:
- The DCG-CNN method successfully balanced sensor fault datasets by enriching small data samples.
- Fault diagnosis accuracy using the DCG-CNN method surpassed traditional approaches.
- Visualization of CNN feature maps provided insights into the model's high performance.
Conclusions:
- The proposed DCG-CNN method offers a robust solution for hydrogen sensor fault diagnosis, particularly with imbalanced data.
- This approach enhances diagnostic accuracy and provides a deeper understanding of the underlying mechanisms.
